font
Filter
×
2026
Deshmukh, Jayati; Yazdanpanah, Vahid; Stein, Sebastian; Ramchurn, Gopal
The triad of identity, trust and responsibility in multi-agent systems Proceedings Article
In: 25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26), 2026.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Coordination, Ethics, Identity, Multiagent Systems, Norms, Responsibility, Trust
@inproceedings{soton509930,
title = {The triad of identity, trust and responsibility in multi-agent systems},
author = {Jayati Deshmukh and Vahid Yazdanpanah and Sebastian Stein and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/509930/},
year = {2026},
date = {2026-05-01},
booktitle = {25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26)},
abstract = {The design of autonomous AI agents that behave responsibly and foster trust in open multi-agent systems remains a fundamental challenge. Traditional game-theoretical approaches largely assume self-interested behaviour, yet real-world collaborations among humans often rely on prosocial considerations that extend beyond individual utility. To address this, for the first time in this paper, we investigate the triad of identity, responsibility, and trust as core elements shaping responsible multi-agent behaviour. We propose a novel agent model, building on the notion of Computational Transcendence, which equips agents with an elastic sense of identity, enabling them to incorporate the welfare of others into their decision-making. Our framework integrates subjective (identity-based) and objective (experience-based and reputation-based) components of trust. Using Iterated Prisoner?s Dilemma (IPD) simulations on different network structures, we analyse how varying levels of identity and trust affect responsible behaviour. Results demonstrate that the interplay of these three concepts can promote emergent responsibility, mitigate exploitation, and sustain long-term cooperation in dynamic multi-agent environments. We argue that this triadic perspective provides a principled foundation for designing trustworthy, responsible, and identity/value aware agents with implications for future human?AI collaboration.},
keywords = {Artificial Intelligence, Coordination, Ethics, Identity, Multiagent Systems, Norms, Responsibility, Trust},
pubstate = {published},
tppubtype = {inproceedings}
}
The design of autonomous AI agents that behave responsibly and foster trust in open multi-agent systems remains a fundamental challenge. Traditional game-theoretical approaches largely assume self-interested behaviour, yet real-world collaborations among humans often rely on prosocial considerations that extend beyond individual utility. To address this, for the first time in this paper, we investigate the triad of identity, responsibility, and trust as core elements shaping responsible multi-agent behaviour. We propose a novel agent model, building on the notion of Computational Transcendence, which equips agents with an elastic sense of identity, enabling them to incorporate the welfare of others into their decision-making. Our framework integrates subjective (identity-based) and objective (experience-based and reputation-based) components of trust. Using Iterated Prisoner?s Dilemma (IPD) simulations on different network structures, we analyse how varying levels of identity and trust affect responsible behaviour. Results demonstrate that the interplay of these three concepts can promote emergent responsibility, mitigate exploitation, and sustain long-term cooperation in dynamic multi-agent environments. We argue that this triadic perspective provides a principled foundation for designing trustworthy, responsible, and identity/value aware agents with implications for future human?AI collaboration.
Georgara, Athina; Deshmukh, Jayati; Ramchurn, Gopal
GreenLine: A delay-tolerant mechanism design for grid capacity allocation Proceedings Article
In: 25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26), pp. 3323–3325, 2026.
Abstract | Links | BibTeX | Tags: auction, mechanism design, power grid, resource allocation
@inproceedings{soton511473,
title = {GreenLine: A delay-tolerant mechanism design for grid capacity allocation},
author = {Athina Georgara and Jayati Deshmukh and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/511473/},
year = {2026},
date = {2026-05-01},
booktitle = {25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26)},
pages = {3323–3325},
abstract = {This work presents GreenLine, a new auction-based mechanism for integrating renewable power plants (RPP) in the power grid. We address a challenging task, where RPP owners want to maximise their profits by installing new power plants in the grid; while, the Distribution Network Operator (DNO) seeks to maximise the generated power while reducing potential installation delays due to growing power demand. This paper formulates this task as a multi-agent system, studies its useful properties such as incentive compatibility, individual rationality and economical efficiency, and discusses GreenLine under different deployment variations.},
keywords = {auction, mechanism design, power grid, resource allocation},
pubstate = {published},
tppubtype = {inproceedings}
}
This work presents GreenLine, a new auction-based mechanism for integrating renewable power plants (RPP) in the power grid. We address a challenging task, where RPP owners want to maximise their profits by installing new power plants in the grid; while, the Distribution Network Operator (DNO) seeks to maximise the generated power while reducing potential installation delays due to growing power demand. This paper formulates this task as a multi-agent system, studies its useful properties such as incentive compatibility, individual rationality and economical efficiency, and discusses GreenLine under different deployment variations.
Singh, Lokesh; Deshmukh, Jayati; Georgara, Athina; Nguyen, Tan Viet Tuyen; Ramchurn, Gopal
CareOps: A multi agent control room for independent living with care Proceedings Article
In: 25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26), pp. 4146–4148, 2026.
Abstract | Links | BibTeX | Tags: Assistive robotics, Human-robot interaction, Independent living, Multi-agent Coordination
@inproceedings{soton511472,
title = {CareOps: A multi agent control room for independent living with care},
author = {Lokesh Singh and Jayati Deshmukh and Athina Georgara and Tan Viet Tuyen Nguyen and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/511472/},
year = {2026},
date = {2026-05-01},
booktitle = {25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26)},
pages = {4146–4148},
abstract = {This paper introduces CareOps, a multi-agent control-room dashboard for Independent Living with Care (ILWC) that coordinates heterogeneous sensors and a Buddy robot across multiple simulated homes. Agents fuse radar, audio, bed, passive infrared (PIR), and gait data into single prioritised incidents with human-readable explanations, and support one-click dispatch of either a robot (via socket) or a human carer (via smartphone notification). The demo presents a single decision-support workflow that orchestrates sensors, software agents, and robots through a unified interface. It provides a human-in-the-loop platform for exploring how professional care staff understand multi-sensor evidence and choose between robot and human responses.},
keywords = {Assistive robotics, Human-robot interaction, Independent living, Multi-agent Coordination},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper introduces CareOps, a multi-agent control-room dashboard for Independent Living with Care (ILWC) that coordinates heterogeneous sensors and a Buddy robot across multiple simulated homes. Agents fuse radar, audio, bed, passive infrared (PIR), and gait data into single prioritised incidents with human-readable explanations, and support one-click dispatch of either a robot (via socket) or a human carer (via smartphone notification). The demo presents a single decision-support workflow that orchestrates sensors, software agents, and robots through a unified interface. It provides a human-in-the-loop platform for exploring how professional care staff understand multi-sensor evidence and choose between robot and human responses.
Hafez, Muhammad Burhan; Miller, Emily; Milford, Michael J.; Ramchurn, Gopal; Ehsan, Shoaib
Through the lens of doubt: robust and efficient uncertainty estimation for visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 11, no. 5, pp. 5899–5906, 2026.
Abstract | Links | BibTeX | Tags: Deep Learning for Visual Perception, Localization, Vision-Based Navigation
@article{soton510726,
title = {Through the lens of doubt: robust and efficient uncertainty estimation for visual place recognition},
author = {Muhammad Burhan Hafez and Emily Miller and Michael J. Milford and Gopal Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/510726/},
year = {2026},
date = {2026-03-01},
journal = {IEEE Robotics and Automation Letters},
volume = {11},
number = {5},
pages = {5899–5906},
abstract = {ensuremath<pensuremath>Visual Place Recognition (VPR) enables robots and autonomous vehicles to identify previously visited locations by matching current observations against a database of known places. However, VPR systems face significant challenges when deployed across varying visual environments, lighting conditions, seasonal changes, and viewpoints changes. Failure-critical VPR applications, such as loop closure detection in simultaneous localization and mapping (SLAM) pipelines, require robust estimation of place matching uncertainty. We propose three training-free uncertainty metrics that estimate prediction confidence by analyzing inherent statistical patterns in similarity scores from any existing VPR method. Similarity Distribution (SD) quantifies match distinctiveness by measuring score separation between candidates; Ratio Spread (RS) evaluates competitive ambiguity among top-scoring locations; and Statistical Uncertainty (SU) is a combination of SD and RS that provides a unified metric that generalizes across datasets and VPR methods without requiring validation data to select the optimal metric. All three metrics operate without additional model training, architectural modifications, or computationally expensive geometric verification. Comprehensive evaluation across nine state-of-the-art VPR methods and six benchmark datasets confirms that our metrics excel at discriminating between correct and incorrect VPR matches, and consistently outperform existing approaches while maintaining negligible computational overhead, making it deployable for real-time robotic applications across varied environmental conditions with improved precision-recall performance.ensuremath</pensuremath>},
keywords = {Deep Learning for Visual Perception, Localization, Vision-Based Navigation},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Visual Place Recognition (VPR) enables robots and autonomous vehicles to identify previously visited locations by matching current observations against a database of known places. However, VPR systems face significant challenges when deployed across varying visual environments, lighting conditions, seasonal changes, and viewpoints changes. Failure-critical VPR applications, such as loop closure detection in simultaneous localization and mapping (SLAM) pipelines, require robust estimation of place matching uncertainty. We propose three training-free uncertainty metrics that estimate prediction confidence by analyzing inherent statistical patterns in similarity scores from any existing VPR method. Similarity Distribution (SD) quantifies match distinctiveness by measuring score separation between candidates; Ratio Spread (RS) evaluates competitive ambiguity among top-scoring locations; and Statistical Uncertainty (SU) is a combination of SD and RS that provides a unified metric that generalizes across datasets and VPR methods without requiring validation data to select the optimal metric. All three metrics operate without additional model training, architectural modifications, or computationally expensive geometric verification. Comprehensive evaluation across nine state-of-the-art VPR methods and six benchmark datasets confirms that our metrics excel at discriminating between correct and incorrect VPR matches, and consistently outperform existing approaches while maintaining negligible computational overhead, making it deployable for real-time robotic applications across varied environmental conditions with improved precision-recall performance.ensuremath</pensuremath>
Tuyen, Nguyen Tan Viet; Georgara, Athina; Singh, Lokesh; Deshmukh, Jayati; Davey, Sean; Tisdale, Paul N.; Ramchurn, Sarvapali
Supporting ?Stop and Watch? in elderly care with socially assistive robots: insights from a participatory design workshop Proceedings Article
In: Baillie, Lynne; Smart, William D.; Graaf, Maartje De; Gombolay, Matthew; Torre, Ilaria (Ed.): Companion Proceedings of the 21st ACM/IEEE International Conference on Human-Robot Interaction, HRI Companion 2026, pp. 302–306, ACM Press, 2026.
Abstract | Links | BibTeX | Tags: assistive living technologies, elderly care, featured_publication, socially assistive robots
@inproceedings{soton511593,
title = {Supporting ?Stop and Watch? in elderly care with socially assistive robots: insights from a participatory design workshop},
author = {Nguyen Tan Viet Tuyen and Athina Georgara and Lokesh Singh and Jayati Deshmukh and Sean Davey and Paul N. Tisdale and Sarvapali Ramchurn},
editor = {Lynne Baillie and William D. Smart and Maartje De Graaf and Matthew Gombolay and Ilaria Torre},
url = {https://eprints.soton.ac.uk/511593/},
year = {2026},
date = {2026-03-01},
urldate = {2026-03-01},
booktitle = {Companion Proceedings of the 21st ACM/IEEE International Conference on Human-Robot Interaction, HRI Companion 2026},
pages = {302–306},
publisher = {ACM Press},
abstract = {ensuremath<pensuremath>?Stop and Watch? is an early-warning tool adopted by the UK NHS and is widely used in elderly care settings. The tool helps caregivers to recognise abnormal changes in residents? health. Despite its clinical value, the process remains highly manual, workload-intensive, and vulnerable to missed observations, particularly in environments facing staff shortages, frequent staff rotations, and increasing care demands. We argue that AI-based systems such as Socially Assistive Robots (SARs) and Assistive Living Technologies (ALTs) offer promising avenues for supporting and enhancing the ?Stop and Watch? tool. However, designing such systems requires a multidisciplinary effort to establish a comprehensive understanding of current practices and the priorities and concerns of all relevant stakeholders. This paper presents insights from a participatory design workshop held in a care home in the UK to explore how SARs and ALTs could meaningfully support the ?Stop and Watch? tool, understand stakeholders? expectations, perceived benefits, and concerns regarding deployment in this sensitive context.ensuremath</pensuremath>},
keywords = {assistive living technologies, elderly care, featured_publication, socially assistive robots},
pubstate = {published},
tppubtype = {inproceedings}
}
ensuremath<pensuremath>?Stop and Watch? is an early-warning tool adopted by the UK NHS and is widely used in elderly care settings. The tool helps caregivers to recognise abnormal changes in residents? health. Despite its clinical value, the process remains highly manual, workload-intensive, and vulnerable to missed observations, particularly in environments facing staff shortages, frequent staff rotations, and increasing care demands. We argue that AI-based systems such as Socially Assistive Robots (SARs) and Assistive Living Technologies (ALTs) offer promising avenues for supporting and enhancing the ?Stop and Watch? tool. However, designing such systems requires a multidisciplinary effort to establish a comprehensive understanding of current practices and the priorities and concerns of all relevant stakeholders. This paper presents insights from a participatory design workshop held in a care home in the UK to explore how SARs and ALTs could meaningfully support the ?Stop and Watch? tool, understand stakeholders? expectations, perceived benefits, and concerns regarding deployment in this sensitive context.ensuremath</pensuremath>
Everett, Gregory Alan
Models and algorithms to optimise team performance in football PhD Thesis
University of Southampton, 2026.
Abstract | Links | BibTeX | Tags: applied machine learning, machine learning, multi-agent systems, sports analytics, team optimisation
@phdthesis{soton509865,
title = {Models and algorithms to optimise team performance in football},
author = {Gregory Alan Everett},
url = {https://eprints.soton.ac.uk/509865/},
year = {2026},
date = {2026-03-01},
publisher = {University of Southampton},
school = {University of Southampton},
abstract = {With the growing use of Artificial Intelligence (AI) and machine learning, there is increasing potential to model and optimise team behaviour across important domains such as disaster response, security, and team sports. These domains are inherently spatiotemporal, requiring models that capture both spatial and temporal dynamics. This thesis focuses on football, a complex, dynamic sport with rich spatiotemporal data and clear objectives, making it an ideal testbed for developing and validating team-based AI models. Moreover, football analytics is a rapidly expanding industry, with European clubs generating ?38 billion in revenue during the 2023/24 season, driving the demand for models that provide competitive advantages.ensuremath<br/ensuremath>ensuremath<br/ensuremath>This thesis proposes a number of novel methods that utilise spatiotemporal data to advance team prediction, analysis, and decision-making in football and other team-based domains. In particular, we introduce a spatiotemporal agent behaviour imputation model that reduces predictive error by 62% compared to baselines in limited observability settings, significantly improving the accessibility of off-ball football analytics through imputed tracking data. We also present a novel spatial teamwork model, combining Monte Carlo tree search (MCTS) and linear programming, that optimises agent decision-making in real-world football defence, reducing opponent threat by 24%. In addition, we develop new metrics, derived from a graph attention network (GAT), to assign credit to indirect agent contributions in team-based defence. The GAT model predicts football passes with a 6% reduction in loss compared to baselines, and we show how these new metrics can greatly improve off-ball football player evaluation. Finally, we propose a dynamic team formation and agent replacement model that accounts for agent fatigue and unavailability and optimises decision-making using a multi-step MCTS algorithm. Applied to football team selection and substitutions, this model improves long-term team performance by 1% and reduces first-team injuries by 15%. This thesis also highlights key remaining challenges for AI in football, including the use of richer player data (e.g., body pose) and greater use of explainable models to build trust in clubs.},
keywords = {applied machine learning, machine learning, multi-agent systems, sports analytics, team optimisation},
pubstate = {published},
tppubtype = {phdthesis}
}
With the growing use of Artificial Intelligence (AI) and machine learning, there is increasing potential to model and optimise team behaviour across important domains such as disaster response, security, and team sports. These domains are inherently spatiotemporal, requiring models that capture both spatial and temporal dynamics. This thesis focuses on football, a complex, dynamic sport with rich spatiotemporal data and clear objectives, making it an ideal testbed for developing and validating team-based AI models. Moreover, football analytics is a rapidly expanding industry, with European clubs generating ?38 billion in revenue during the 2023/24 season, driving the demand for models that provide competitive advantages.ensuremath<br/ensuremath>ensuremath<br/ensuremath>This thesis proposes a number of novel methods that utilise spatiotemporal data to advance team prediction, analysis, and decision-making in football and other team-based domains. In particular, we introduce a spatiotemporal agent behaviour imputation model that reduces predictive error by 62% compared to baselines in limited observability settings, significantly improving the accessibility of off-ball football analytics through imputed tracking data. We also present a novel spatial teamwork model, combining Monte Carlo tree search (MCTS) and linear programming, that optimises agent decision-making in real-world football defence, reducing opponent threat by 24%. In addition, we develop new metrics, derived from a graph attention network (GAT), to assign credit to indirect agent contributions in team-based defence. The GAT model predicts football passes with a 6% reduction in loss compared to baselines, and we show how these new metrics can greatly improve off-ball football player evaluation. Finally, we propose a dynamic team formation and agent replacement model that accounts for agent fatigue and unavailability and optimises decision-making using a multi-step MCTS algorithm. Applied to football team selection and substitutions, this model improves long-term team performance by 1% and reduces first-team injuries by 15%. This thesis also highlights key remaining challenges for AI in football, including the use of richer player data (e.g., body pose) and greater use of explainable models to build trust in clubs.
Ramchurn, Gopal; Neff, Gina; Parisio, Isabela; Kiden, Sarah; Georgara, Athina; Stumpf, Simone; Wong, Mark; Shahandashti, Siamak F.; Aristodemou, Marios
Responsible AI UK response to the DSIT?s call for evidence on the AI Growth Lab Technical Report
no. 10.5258/RAi/005, 2026.
Abstract | Links | BibTeX | Tags:
@techreport{soton508020,
title = {Responsible AI UK response to the DSIT?s call for evidence on the AI Growth Lab},
author = {Gopal Ramchurn and Gina Neff and Isabela Parisio and Sarah Kiden and Athina Georgara and Simone Stumpf and Mark Wong and Siamak F. Shahandashti and Marios Aristodemou},
url = {https://eprints.soton.ac.uk/508020/},
year = {2026},
date = {2026-01-01},
number = {10.5258/RAi/005},
publisher = {University of Southampton},
abstract = {A response to the Department for Science, Innovation and Technology (DSIT) open call for evidence regarding the AI Growth Lab1 on behalf of Responsible AI UK (RAi UK), an open and multidisciplinary network that brings together experts from across the four nations of the UK to understand how we should shape the development of AI to benefit people, communities and society},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
A response to the Department for Science, Innovation and Technology (DSIT) open call for evidence regarding the AI Growth Lab1 on behalf of Responsible AI UK (RAi UK), an open and multidisciplinary network that brings together experts from across the four nations of the UK to understand how we should shape the development of AI to benefit people, communities and society
2025
Reyes-Cruz, Gisela; Kiden, Sarah; Azim, Tayyaba; Bergin, Aislinn Gomez; Choi, Sena; Eke, Damian; Klyshbekova, Maira; Peter, Oriane; Waheed, Maria; Devlin, Kate; Fischer, Joel; Vallejos, Elvira Perez; Ramchurn, Gopal; Stein, Sebastian
Response to call for input for EMRTD study ?Artificial Intelligence, Cultural Rights, and the Right to Development? Technical Report
no. 10.25878/7nmj-0t79, 2025.
Abstract | Links | BibTeX | Tags:
@techreport{soton511244,
title = {Response to call for input for EMRTD study ?Artificial Intelligence, Cultural Rights, and the Right to Development?},
author = {Gisela Reyes-Cruz and Sarah Kiden and Tayyaba Azim and Aislinn Gomez Bergin and Sena Choi and Damian Eke and Maira Klyshbekova and Oriane Peter and Maria Waheed and Kate Devlin and Joel Fischer and Elvira Perez Vallejos and Gopal Ramchurn and Sebastian Stein},
url = {https://eprints.soton.ac.uk/511244/},
year = {2025},
date = {2025-11-01},
number = {10.25878/7nmj-0t79},
publisher = {University of Nottingham},
abstract = {Responsible AI UK's response to the Call for Input for EMRTD study "Artificial Intelligence, Cultural Rights, and the Right to Development" issued by the Expert Mechanism on the Right to Development United Nations Human Rights Office of the High Commissioner (OHCHR).ensuremath<br/ensuremath>ensuremath<br/ensuremath>https://www.ohchr.org/en/calls-for-input/2025/call-input-emrtd-study-artificial-intelligence-cultural-rights-and-right},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
Responsible AI UK's response to the Call for Input for EMRTD study "Artificial Intelligence, Cultural Rights, and the Right to Development" issued by the Expert Mechanism on the Right to Development United Nations Human Rights Office of the High Commissioner (OHCHR).ensuremath<br/ensuremath>ensuremath<br/ensuremath>https://www.ohchr.org/en/calls-for-input/2025/call-input-emrtd-study-artificial-intelligence-cultural-rights-and-right
Owen, Aled Lloyd; Ramchurn, Gopal; Dasgupta, Prokar; Ao, Shuang; Barnard, Pepita; Deshmukh, Jayati; Parisio, Isabela; Singh, Lokesh; Valoor, Adarsh; Waheed, Maria; Braunschweig, Bertrand; Quintero, Karla; Poretschkin, Maximilian; Hawes, Ben; Khomh, Foutse
Advancing Trustworthy Artificial Intelligence: Lessons Learned and Emerging Challenges Proceedings Article
In: Responsible Ai, University of Southampton, 2025.
Abstract | Links | BibTeX | Tags:
@inproceedings{soton506660,
title = {Advancing Trustworthy Artificial Intelligence: Lessons Learned and Emerging Challenges},
author = {Aled Lloyd Owen and Gopal Ramchurn and Prokar Dasgupta and Shuang Ao and Pepita Barnard and Jayati Deshmukh and Isabela Parisio and Lokesh Singh and Adarsh Valoor and Maria Waheed and Bertrand Braunschweig and Karla Quintero and Maximilian Poretschkin and Ben Hawes and Foutse Khomh},
url = {https://eprints.soton.ac.uk/506660/},
year = {2025},
date = {2025-11-01},
booktitle = {Responsible Ai},
publisher = {University of Southampton},
abstract = {Government white paper},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Government white paper
Kiden, Sarah; Peter, Oriane; Reyes-Cruz, Gisela; Klyshbekova, Maira; Choi, Sena; Bergin, Aislinn Gomez; Waheed, Maria; Eke, Damian; Azim, Tayyaba; Ramchurn, Sarvapali; Stein, Sebastian; Vallejos, Elvira Perez; Devlin, Kate; Fischer, Joel E.
Back to the communities: a mixed-methods and community-driven evaluation of cultural sensitivity in text-to-image models Miscellaneous
2025.
Abstract | Links | BibTeX | Tags: cs.CY, cs.SI
@misc{soton507303,
title = {Back to the communities: a mixed-methods and community-driven evaluation of cultural sensitivity in text-to-image models},
author = {Sarah Kiden and Oriane Peter and Gisela Reyes-Cruz and Maira Klyshbekova and Sena Choi and Aislinn Gomez Bergin and Maria Waheed and Damian Eke and Tayyaba Azim and Sarvapali Ramchurn and Sebastian Stein and Elvira Perez Vallejos and Kate Devlin and Joel E. Fischer},
url = {https://eprints.soton.ac.uk/507303/},
year = {2025},
date = {2025-10-01},
publisher = {arXiv},
abstract = {Evidence shows that text-to-image (T2I) models disproportionately reflect Western cultural norms, amplifying misrepresentation and harms to minority groups. However, evaluating cultural sensitivity is inherently complex due to its fluid and multifaceted nature. This paper draws on a state-of-the-art review and co-creation workshops involving 59 individuals from 19 different countries. We developed and validated a mixed-methods community-based evaluation methodology to assess cultural sensitivity in T2I models, which embraces first-person methods. Quantitative scores and qualitative inquiries expose convergence and disagreement within and across communities, illuminate the downstream consequences of misrepresentation, and trace how training data shaped by unequal power relations distort depictions. Extensive assessments are constrained by high resource requirements and the dynamic nature of culture, a tension we alleviate through a context-based and iterative methodology. The paper provides actionable recommendations for stakeholders, highlighting pathways to investigate the sources, mechanisms, and impacts of cultural (mis)representation in T2I models.},
keywords = {cs.CY, cs.SI},
pubstate = {published},
tppubtype = {misc}
}
Evidence shows that text-to-image (T2I) models disproportionately reflect Western cultural norms, amplifying misrepresentation and harms to minority groups. However, evaluating cultural sensitivity is inherently complex due to its fluid and multifaceted nature. This paper draws on a state-of-the-art review and co-creation workshops involving 59 individuals from 19 different countries. We developed and validated a mixed-methods community-based evaluation methodology to assess cultural sensitivity in T2I models, which embraces first-person methods. Quantitative scores and qualitative inquiries expose convergence and disagreement within and across communities, illuminate the downstream consequences of misrepresentation, and trace how training data shaped by unequal power relations distort depictions. Extensive assessments are constrained by high resource requirements and the dynamic nature of culture, a tension we alleviate through a context-based and iterative methodology. The paper provides actionable recommendations for stakeholders, highlighting pathways to investigate the sources, mechanisms, and impacts of cultural (mis)representation in T2I models.
Ramchurn, Gopal; Owen, Aled Lloyd; Ao, Shuang; Barnard, Pepita; Bergin, Aislinn Gomez; Parisio, Isabela; Valoor, Adarsh; Waheed, Maria; Holter, Carolyn Ten; Portillo, Virginia; Procter, Rob; Batterham, Paul; Downer, John; Winter, Peter; Krook, Joshua; Blockx, Jan; Braunschweig, Bertrand; Quintero, Karla; Poretschkin, Maximilian; Becker, Daniel; Campbell-Ratcliffe, Emily; Hawes, Ben; Shaw, Patricia; Patel, Reema; Tewari, Ashish; Thomas, Alec; Duncan, Paul; Gillings, Eliot
Frameworks and Toolkits for Assuring Responsible AI Book Section
In: Responsible Ai UK, University of Southampton, 2025.
Abstract | Links | BibTeX | Tags: featured_publication
@incollection{soton506057,
title = {Frameworks and Toolkits for Assuring Responsible AI},
author = {Gopal Ramchurn and Aled Lloyd Owen and Shuang Ao and Pepita Barnard and Aislinn Gomez Bergin and Isabela Parisio and Adarsh Valoor and Maria Waheed and Carolyn Ten Holter and Virginia Portillo and Rob Procter and Paul Batterham and John Downer and Peter Winter and Joshua Krook and Jan Blockx and Bertrand Braunschweig and Karla Quintero and Maximilian Poretschkin and Daniel Becker and Emily Campbell-Ratcliffe and Ben Hawes and Patricia Shaw and Reema Patel and Ashish Tewari and Alec Thomas and Paul Duncan and Eliot Gillings},
url = {https://eprints.soton.ac.uk/506057/},
year = {2025},
date = {2025-10-01},
urldate = {2025-10-01},
booktitle = {Responsible Ai UK},
publisher = {University of Southampton},
abstract = {This report explores experiences of using tools for assuring responsible practice in the build, deployment, and governance of AI systems, in particular frameworks and toolkits developed and published by RAi UK-funded projects and by other organisations. We examine evidence of how and why these tools are used, and of their usefulness and their limitations. We look for lessons that could improve AI assurance in the future, including potentially using AI-based tools to support AI assurance. The aim is to identify priority areas and key questions for RAi UK and other researchers and organisations to explore further, with the goal of improving the fit between the supply of, and demand for, tools that support organisational assurance of responsible AI .Responsible Ai UK and our partner for this event, Confiance.ai, brought people together from across government, public services, and industry, as well as project teams that develop support for responsible AI. The aim was to clarify how people should approach the development and use of toolkits and/or frameworks in practice. We wanted to synthesise findings from our research across the programme and to find and address any gaps. The workshop took place online on Thursday 10 April 2025.},
keywords = {featured_publication},
pubstate = {published},
tppubtype = {incollection}
}
This report explores experiences of using tools for assuring responsible practice in the build, deployment, and governance of AI systems, in particular frameworks and toolkits developed and published by RAi UK-funded projects and by other organisations. We examine evidence of how and why these tools are used, and of their usefulness and their limitations. We look for lessons that could improve AI assurance in the future, including potentially using AI-based tools to support AI assurance. The aim is to identify priority areas and key questions for RAi UK and other researchers and organisations to explore further, with the goal of improving the fit between the supply of, and demand for, tools that support organisational assurance of responsible AI .Responsible Ai UK and our partner for this event, Confiance.ai, brought people together from across government, public services, and industry, as well as project teams that develop support for responsible AI. The aim was to clarify how people should approach the development and use of toolkits and/or frameworks in practice. We wanted to synthesise findings from our research across the programme and to find and address any gaps. The workshop took place online on Thursday 10 April 2025.
Thavanesan, Navamayooran; Naiseh, Mohammad; Terol, Miguel; Rahman, Saqib Andrew; Hill, Samuel Luke; Parfitt, Charlotte; Walters, Zoë S; Ramchurn, Sarvapali; Markar, Sheraz; Owen, Richard; Maynard, Nick; Azim, Tayyaba; Belkatir, Zehor; Perez, Elvira Vallejos; McCord, Mimi; Underwood, Tim; Vigneswaran, Ganesh
The oesophageal cancer multi-disciplinary tool: a co-designed, externally validated, machine learning clinical decision support system Journal Article
In: EClinicalMedicine, vol. 89, pp. 103527, 2025, (For the purposes of open access, the authors have applied a Creative Commons attribution license (CC-BY) to any Author Accepted Manuscript version arising from this submission.).
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Decision support tool, machine learning, MDT, Multidisciplinary teams, Oesophageal cancer
@article{soton506614,
title = {The oesophageal cancer multi-disciplinary tool: a co-designed, externally validated, machine learning clinical decision support system},
author = {Navamayooran Thavanesan and Mohammad Naiseh and Miguel Terol and Saqib Andrew Rahman and Samuel Luke Hill and Charlotte Parfitt and Zoë S Walters and Sarvapali Ramchurn and Sheraz Markar and Richard Owen and Nick Maynard and Tayyaba Azim and Zehor Belkatir and Elvira Vallejos Perez and Mimi McCord and Tim Underwood and Ganesh Vigneswaran},
url = {https://eprints.soton.ac.uk/506614/},
year = {2025},
date = {2025-09-01},
journal = {EClinicalMedicine},
volume = {89},
pages = {103527},
abstract = {ensuremath<pensuremath>BACKGROUND: The oesophageal cancer (OC) multi-disciplinary team (MDT) operates under significant pressures, handling complex decision-making. Machine learning (ML) can learn complex decision-making paradigms to improve efficiency, consistency, and cost if trained and deployed responsibly. We present an externally validated ML-based clinical decision support system (CDSS) designed to predict OC MDT treatment decisions and prognosticate palliative scenarios, co-designed using Responsible Research and Innovation (RRI) principles.ensuremath</pensuremath>ensuremath<pensuremath>METHODS: Clinicopathological data collected from 1931 patients between 4th September 2009, and 8th November 2022 were used to test and validate models trained through four ML algorithms to predict curative and palliative treatment pathways along with palliative prognosis. 953 OC cases treated at University Hospitals Southampton (UHS) were used to train ML models which were externally validated on 978 OC cases from Oxford University Hospitals (OUH). Model performance was evaluated using Area Under Curve (AUC) for treatment classifiers and calibration curves for survival models. A parallel RRI program at the University of Southampton (United Kingdom) combining clinician interviews and inter-disciplinary workshops was conducted between 16.3.23 and 23.5.24. The RRI program comprised a group of 17 domain experts comprising programmers, computer scientists, clinicians and patient representatives to allow end-users to contribute towards the co-design of the CDSS user interface.ensuremath</pensuremath>ensuremath<pensuremath>FINDINGS: Cohorts differed in baseline characteristics, with the external cohort (OUH) being younger, having better performance status, and a higher prevalence of pulmonary and vascular disease. Despite these differences, on internal validation (UHS cohort) mean AUCs for the primary treatment model were: MLR 0.905 $±$ 0.048, XGB 0.909 $±$ 0.044 and RF 0.883 $±$ 0.059 (k = 5 cross-validation) and MLR 0.866 (95% CI 0.866-0.867), XGB 0.863 (0.862-0.864), RF 0.863 (0.867-0.868) on bootstrapped resampling. For the palliative classifier, mean AUCs were: MLR 0.805 $±$ 0.096, XGB 0.815 $±$ 0.081 and RF 0.793 $±$ 0.083 (k = 5 cross-validation) and MLR 0.736 (95% CI 0.734-0.737), XGB 0.799 (0.798-0.800), RF 0.781 (0.778-0.782) on bootstrapped resampling. On external validation (OUH cohort), AUCs were MLR 0.894, XGB 0.887 and RF 0.891 for the primary treatment model and MLR 0.711, XGB 0.742 and RF 0.730 for the palliative treatment classifier. Predicted survival probability from the palliative survival model was well calibrated over the first 12 months post-diagnosis in both cohorts. The RRI program provided a collaborative environment leading to valuable modifications to the CDSS including prediction explanations, visual aids for survival and integrated education for users producing a user-friendly and quick to use tool.ensuremath</pensuremath>ensuremath<pensuremath>INTERPRETATION: We present a novel, responsibly developed, externally validated AI CDSS trained to predict oesophageal cancer MDT decisions. It represents the foundations of a transformative application of ML, personalised, consistent and efficient MDT decision-support within OC which aligns to RRI principles.ensuremath</pensuremath>ensuremath<pensuremath>FUNDING: Doctoral Studentship for NT (Institute for Life Sciences (University of Southampton) & University Hospital Southampton), UKRI TAS Pump-Priming Grant (TAS_PP_00167).ensuremath</pensuremath>},
note = {For the purposes of open access, the authors have applied a Creative Commons attribution license (CC-BY) to any Author Accepted Manuscript version arising from this submission.},
keywords = {Artificial Intelligence, Decision support tool, machine learning, MDT, Multidisciplinary teams, Oesophageal cancer},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>BACKGROUND: The oesophageal cancer (OC) multi-disciplinary team (MDT) operates under significant pressures, handling complex decision-making. Machine learning (ML) can learn complex decision-making paradigms to improve efficiency, consistency, and cost if trained and deployed responsibly. We present an externally validated ML-based clinical decision support system (CDSS) designed to predict OC MDT treatment decisions and prognosticate palliative scenarios, co-designed using Responsible Research and Innovation (RRI) principles.ensuremath</pensuremath>ensuremath<pensuremath>METHODS: Clinicopathological data collected from 1931 patients between 4th September 2009, and 8th November 2022 were used to test and validate models trained through four ML algorithms to predict curative and palliative treatment pathways along with palliative prognosis. 953 OC cases treated at University Hospitals Southampton (UHS) were used to train ML models which were externally validated on 978 OC cases from Oxford University Hospitals (OUH). Model performance was evaluated using Area Under Curve (AUC) for treatment classifiers and calibration curves for survival models. A parallel RRI program at the University of Southampton (United Kingdom) combining clinician interviews and inter-disciplinary workshops was conducted between 16.3.23 and 23.5.24. The RRI program comprised a group of 17 domain experts comprising programmers, computer scientists, clinicians and patient representatives to allow end-users to contribute towards the co-design of the CDSS user interface.ensuremath</pensuremath>ensuremath<pensuremath>FINDINGS: Cohorts differed in baseline characteristics, with the external cohort (OUH) being younger, having better performance status, and a higher prevalence of pulmonary and vascular disease. Despite these differences, on internal validation (UHS cohort) mean AUCs for the primary treatment model were: MLR 0.905 $±$ 0.048, XGB 0.909 $±$ 0.044 and RF 0.883 $±$ 0.059 (k = 5 cross-validation) and MLR 0.866 (95% CI 0.866-0.867), XGB 0.863 (0.862-0.864), RF 0.863 (0.867-0.868) on bootstrapped resampling. For the palliative classifier, mean AUCs were: MLR 0.805 $±$ 0.096, XGB 0.815 $±$ 0.081 and RF 0.793 $±$ 0.083 (k = 5 cross-validation) and MLR 0.736 (95% CI 0.734-0.737), XGB 0.799 (0.798-0.800), RF 0.781 (0.778-0.782) on bootstrapped resampling. On external validation (OUH cohort), AUCs were MLR 0.894, XGB 0.887 and RF 0.891 for the primary treatment model and MLR 0.711, XGB 0.742 and RF 0.730 for the palliative treatment classifier. Predicted survival probability from the palliative survival model was well calibrated over the first 12 months post-diagnosis in both cohorts. The RRI program provided a collaborative environment leading to valuable modifications to the CDSS including prediction explanations, visual aids for survival and integrated education for users producing a user-friendly and quick to use tool.ensuremath</pensuremath>ensuremath<pensuremath>INTERPRETATION: We present a novel, responsibly developed, externally validated AI CDSS trained to predict oesophageal cancer MDT decisions. It represents the foundations of a transformative application of ML, personalised, consistent and efficient MDT decision-support within OC which aligns to RRI principles.ensuremath</pensuremath>ensuremath<pensuremath>FUNDING: Doctoral Studentship for NT (Institute for Life Sciences (University of Southampton) & University Hospital Southampton), UKRI TAS Pump-Priming Grant (TAS_PP_00167).ensuremath</pensuremath>
Ansari, Aamir Ahmad; Ramchurn, Gopal; Nguyen, Tan Viet Tuyen
Beyond text: multi-modal LLM in human robot interaction Proceedings Article
In: UK AI Research Symposium: A Festival of Ideas (08/09/25 - 09/09/25), 2025.
Abstract | Links | BibTeX | Tags:
@inproceedings{soton505659,
title = {Beyond text: multi-modal LLM in human robot interaction},
author = {Aamir Ahmad Ansari and Gopal Ramchurn and Tan Viet Tuyen Nguyen},
url = {https://eprints.soton.ac.uk/505659/},
year = {2025},
date = {2025-09-01},
booktitle = {UK AI Research Symposium: A Festival of Ideas (08/09/25 - 09/09/25)},
abstract = {Multimodal interaction plays a vital role in Human-Robot Interaction (HRI), enabling robots to communicate with humans through multiple channels. This study introduces a novel approach to enhance such interactions by treating images and human motion as distinct foreign languages, in addition to text. In the proposed framework, vector quantization is employed to convert multimodal inputs such as images and human motions to an aligned set of tokens. A Large Language Model (LLM) is then pre-trained with the use of Low-Rank Adaptation (LoRA) and instruction-tuned on a dialogue dataset that incorporates both image and motion context. The proposed multimodal LLM framework aims to equip robots with the ability to understand and respond to complex human queries through multimodal inputs and outputs, enabling more natural and effective interactions.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Multimodal interaction plays a vital role in Human-Robot Interaction (HRI), enabling robots to communicate with humans through multiple channels. This study introduces a novel approach to enhance such interactions by treating images and human motion as distinct foreign languages, in addition to text. In the proposed framework, vector quantization is employed to convert multimodal inputs such as images and human motions to an aligned set of tokens. A Large Language Model (LLM) is then pre-trained with the use of Low-Rank Adaptation (LoRA) and instruction-tuned on a dialogue dataset that incorporates both image and motion context. The proposed multimodal LLM framework aims to equip robots with the ability to understand and respond to complex human queries through multimodal inputs and outputs, enabling more natural and effective interactions.
Le, Thien Doanh; Nguyen, Tan Viet Tuyen; Duy, Tan Le; Ramchurn, Gopal
A multimodal large language model framework for gesture generation in social robots Proceedings Article
In: BEAR Workshop, The 2025 IEEE International Conference on Robot and Human Interactive Communication (25/08/25 - 29/08/25), 2025.
Abstract | Links | BibTeX | Tags:
@inproceedings{soton506590,
title = {A multimodal large language model framework for gesture generation in social robots},
author = {Thien Doanh Le and Tan Viet Tuyen Nguyen and Tan Le Duy and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/506590/},
year = {2025},
date = {2025-08-01},
booktitle = {BEAR Workshop, The 2025 IEEE International Conference on Robot and Human Interactive Communication (25/08/25 - 29/08/25)},
abstract = {Non-verbal gestures play a crucial role in social robots, enabling them to signal their intentions to users during human?robot interaction (HRI). While recent research in this domain has primarily focused on robot gesture generation, there remains a limited number of studies on multimodal generation frameworks, where generated gestures are harmonized with other generated modalities, to better convey the robot?s intention to users via a wider range of communication channels. Inspired by recent advancements in multimodal large language models (MLLMs), we propose a novel framework that integrates motion generation models with existing MLLMs to produce high-quality 3D motions without the need for extensive multimodal training. Our framework comprises three key components: a Denoising Diffusion Motion Generation (DDMG) module that maps text descriptions to motion sequences using a diffusion-based approach; a Motion Decoding Alignment (MDA) module that refines motion representations by incorporating signal embeddings generated by an LLM; and a Fusion Module (FM) that integrates motion features trained from previous phases to enhance coherence and realism. We conducted a series of experiments on a publicly available dataset to evaluate the efficiency of the proposed framework in terms of motion quality, diversity, and semantic alignment. The results suggest that our multimodal approach can serve as a powerful controller for robot gesture generation, offering a more scalable and effective solution, particularly for social HRI.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Non-verbal gestures play a crucial role in social robots, enabling them to signal their intentions to users during human?robot interaction (HRI). While recent research in this domain has primarily focused on robot gesture generation, there remains a limited number of studies on multimodal generation frameworks, where generated gestures are harmonized with other generated modalities, to better convey the robot?s intention to users via a wider range of communication channels. Inspired by recent advancements in multimodal large language models (MLLMs), we propose a novel framework that integrates motion generation models with existing MLLMs to produce high-quality 3D motions without the need for extensive multimodal training. Our framework comprises three key components: a Denoising Diffusion Motion Generation (DDMG) module that maps text descriptions to motion sequences using a diffusion-based approach; a Motion Decoding Alignment (MDA) module that refines motion representations by incorporating signal embeddings generated by an LLM; and a Fusion Module (FM) that integrates motion features trained from previous phases to enhance coherence and realism. We conducted a series of experiments on a publicly available dataset to evaluate the efficiency of the proposed framework in terms of motion quality, diversity, and semantic alignment. The results suggest that our multimodal approach can serve as a powerful controller for robot gesture generation, offering a more scalable and effective solution, particularly for social HRI.
Ramchurn, Gopal; Jones, Matt; Owen, Aled Lloyd; Ehsan, Shoaib; Kiden, Sarah; Eke, Damian; McStay, Andrew; Reitmaier, Thomas; Raju, Dani Kalarikalayil; Castenada, Nestor; Sailaja, Neelima; Faith, Becky; Hermann, Antony; Kalra, Kanika; Hawes, Ben; Adams, Rachel; Meier, Patrick; Sharma, Gaurav; Vishwarupe, Varad
Responsible AI. To enable flourishing in and by Low- and Middle- Income Countries (LMICs) Book Section
In: Responsible Ai UK, University of Southampton, 2025.
Abstract | Links | BibTeX | Tags:
@incollection{soton506045,
title = {Responsible AI. To enable flourishing in and by Low- and Middle- Income Countries (LMICs)},
author = {Gopal Ramchurn and Matt Jones and Aled Lloyd Owen and Shoaib Ehsan and Sarah Kiden and Damian Eke and Andrew McStay and Thomas Reitmaier and Dani Kalarikalayil Raju and Nestor Castenada and Neelima Sailaja and Becky Faith and Antony Hermann and Kanika Kalra and Ben Hawes and Rachel Adams and Patrick Meier and Gaurav Sharma and Varad Vishwarupe},
url = {https://eprints.soton.ac.uk/506045/},
year = {2025},
date = {2025-07-01},
booktitle = {Responsible Ai UK},
publisher = {University of Southampton},
abstract = {This report explores how governments and industry in Low- and Middle-Income Countries (LMICs) might adopt AI to serve the values, needs and aspirations of individuals and communities, and drive the achievement of development goals. It draws on RAi UK projects working with partners in India and Indonesia and in countries in Africa and in South America, and identifies challenges, early solutions and policy implications. The aim is to identify priority areas and questions for RAi UK and other researchers and organisations to explore further. In this report we focus on the emerging challenges and other outputs of research on AI for LMICs. We convened leaders and members of projects that explore embedding AI tools and resources that are sensitive to local, cultural norms and end-user experience through co-creation with communities.},
keywords = {},
pubstate = {published},
tppubtype = {incollection}
}
This report explores how governments and industry in Low- and Middle-Income Countries (LMICs) might adopt AI to serve the values, needs and aspirations of individuals and communities, and drive the achievement of development goals. It draws on RAi UK projects working with partners in India and Indonesia and in countries in Africa and in South America, and identifies challenges, early solutions and policy implications. The aim is to identify priority areas and questions for RAi UK and other researchers and organisations to explore further. In this report we focus on the emerging challenges and other outputs of research on AI for LMICs. We convened leaders and members of projects that explore embedding AI tools and resources that are sensitive to local, cultural norms and end-user experience through co-creation with communities.
Abioye, Ayodeji O.; Hunt, William; Gu, Yue; Schneiders, Eike; Naiseh, Mohammad; Archibald, Blair; Sevegnani, Michele; Ramchurn, Sarvapali D.; Fischer, Joel E.; Soorati, Mohammad D.
A user study evaluation of predictive formal modelling at runtime in human-swarm interaction Journal Article
In: ACM Transactions on Human-Robot Interaction, vol. 14, no. 4, 2025.
Abstract | Links | BibTeX | Tags: Additional Key Words and PhrasesHuman-Robot Interaction (HRI), Human-Swarm Interaction (HSI), Predictive Formal Modelling (PFM), Usability, Workload
@article{soton501672,
title = {A user study evaluation of predictive formal modelling at runtime in human-swarm interaction},
author = {Ayodeji O. Abioye and William Hunt and Yue Gu and Eike Schneiders and Mohammad Naiseh and Blair Archibald and Michele Sevegnani and Sarvapali D. Ramchurn and Joel E. Fischer and Mohammad D. Soorati},
url = {https://eprints.soton.ac.uk/501672/},
year = {2025},
date = {2025-06-01},
journal = {ACM Transactions on Human-Robot Interaction},
volume = {14},
number = {4},
abstract = {Formal modelling is often used as part of the design and testing process of software development to ensure that components operate within suitable bounds even in unexpected circumstances. We conducted a user study evaluation of predictive formal modelling (PFM) at runtime in a human-swarm mission to determine the benefit of predictive formal modelling on performance and human-swarm interaction. 180 participants were recruited to perform the role of aerial swarm operators delivering parcels to target locations in a simulation environment. The PFM model was integrated into the simulation software to inform the operator of the estimated mission completion time given the current number of drones deployed. The operator could increase the number of parcels delivered in any time step by adding drones, which also increased costs, thus requiring the use of the minimum number of drones necessary to complete the task in the given time. We collected user feedback using standard survey questionnaires and measured performance using data obtained from the Human And Robot Interactive Swarm (HARIS) simulator. Our results show that PFM increased the performance of the human swarm team without significantly increasing the operators? workload or affecting the system's usability.},
keywords = {Additional Key Words and PhrasesHuman-Robot Interaction (HRI), Human-Swarm Interaction (HSI), Predictive Formal Modelling (PFM), Usability, Workload},
pubstate = {published},
tppubtype = {article}
}
Formal modelling is often used as part of the design and testing process of software development to ensure that components operate within suitable bounds even in unexpected circumstances. We conducted a user study evaluation of predictive formal modelling (PFM) at runtime in a human-swarm mission to determine the benefit of predictive formal modelling on performance and human-swarm interaction. 180 participants were recruited to perform the role of aerial swarm operators delivering parcels to target locations in a simulation environment. The PFM model was integrated into the simulation software to inform the operator of the estimated mission completion time given the current number of drones deployed. The operator could increase the number of parcels delivered in any time step by adding drones, which also increased costs, thus requiring the use of the minimum number of drones necessary to complete the task in the given time. We collected user feedback using standard survey questionnaires and measured performance using data obtained from the Human And Robot Interactive Swarm (HARIS) simulator. Our results show that PFM increased the performance of the human swarm team without significantly increasing the operators? workload or affecting the system's usability.
Deshmukh, Jayati; Liang, Zijie; Yazdanpanah, Vahid; Stein, Sebastian; Ramchurn, Savapali D.
Serious games for ethical preference elicitation Proceedings Article
In: AAMAS '25: Proceedings of the 24th International Conference on Autonomous Agents and Multiagent Systems, pp. 3018 – 3020, Association for Computing Machinery, 2025.
Abstract | Links | BibTeX | Tags:
@inproceedings{soton498743b,
title = {Serious games for ethical preference elicitation},
author = {Jayati Deshmukh and Zijie Liang and Vahid Yazdanpanah and Sebastian Stein and Savapali D. Ramchurn},
url = {https://eprints.soton.ac.uk/498743/},
year = {2025},
date = {2025-06-01},
booktitle = {AAMAS '25: Proceedings of the 24th International Conference on Autonomous Agents and Multiagent Systems},
pages = {3018 – 3020},
publisher = {Association for Computing Machinery},
abstract = {Autonomous agents acting on behalf of humans must act according to their ethical preferences. However, ethical preferences are latent and abstract and thus it is challenging to elicit them. To address this, we present a serious game that helps elicit ethical preferences in a more dynamic and engaging way than traditional methods such as questionnaires or simple dilemmas.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Autonomous agents acting on behalf of humans must act according to their ethical preferences. However, ethical preferences are latent and abstract and thus it is challenging to elicit them. To address this, we present a serious game that helps elicit ethical preferences in a more dynamic and engaging way than traditional methods such as questionnaires or simple dilemmas.
Deshmukh, Jayati; Liang, Zijie; Yazdanpanah, Vahid; Stein, Sebastian; Ramchurn, Sarvalpali D.
Serious games for ethical preference elicitation Proceedings Article
In: AAMAS - 2025 : The 24th International Conference on Autonomous Agents and Multiagent Systems (19/05/25 - 23/05/25), 2025.
Abstract | Links | BibTeX | Tags:
@inproceedings{soton498743,
title = {Serious games for ethical preference elicitation},
author = {Jayati Deshmukh and Zijie Liang and Vahid Yazdanpanah and Sebastian Stein and Sarvalpali D. Ramchurn},
url = {https://eprints.soton.ac.uk/498743/},
year = {2025},
date = {2025-05-01},
booktitle = {AAMAS - 2025 : The 24th International Conference on Autonomous Agents and Multiagent Systems (19/05/25 - 23/05/25)},
abstract = {Autonomous agents acting on behalf of humans must act according to their ethical preferences. However, ethical preferences are latent and abstract and thus it is challenging to elicit them. To address this, we present a serious game that helps elicit ethical preferences in a more dynamic and engaging way than traditional methods such as questionnaires or simple dilemmas.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Autonomous agents acting on behalf of humans must act according to their ethical preferences. However, ethical preferences are latent and abstract and thus it is challenging to elicit them. To address this, we present a serious game that helps elicit ethical preferences in a more dynamic and engaging way than traditional methods such as questionnaires or simple dilemmas.
Abioye, Ayodeji; Hunt, William; Schneiders, Eike; Gu, Yue; Naiseh, Mohammad; Archibald, Blair; Sevegnani, Michele; Ramchurn, Gopal; Fischer, Joel; Soorati, Mohammad
Research Data for Paper: A User Study Evaluation of Predictive Formal Modelling at Runtime in Human-Swarm Interaction Miscellaneous
2025.
Abstract | Links | BibTeX | Tags:
@misc{soton500788,
title = {Research Data for Paper: A User Study Evaluation of Predictive Formal Modelling at Runtime in Human-Swarm Interaction},
author = {Ayodeji Abioye and William Hunt and Eike Schneiders and Yue Gu and Mohammad Naiseh and Blair Archibald and Michele Sevegnani and Gopal Ramchurn and Joel Fischer and Mohammad Soorati},
url = {https://eprints.soton.ac.uk/500788/},
year = {2025},
date = {2025-03-01},
publisher = {figshare},
abstract = {ensuremath<span style='white-space:pre-line'ensuremath>This dataset is for the user study conducted to evaluate the performance benefits of predictive formal modelling (PFM) at runtime in a human-swarm interaction experiment. We recruited 180 participants to perform the role of aerial swarm operators conducting a drone delivery mission in a simulation environment using the Human And Robot Interactive Swarm (HARIS) simulator.ensuremath</spanensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
ensuremath<span style='white-space:pre-line'ensuremath>This dataset is for the user study conducted to evaluate the performance benefits of predictive formal modelling (PFM) at runtime in a human-swarm interaction experiment. We recruited 180 participants to perform the role of aerial swarm operators conducting a drone delivery mission in a simulation environment using the Human And Robot Interactive Swarm (HARIS) simulator.ensuremath</spanensuremath>
Ismagilov, Timur; Ferrarini, Bruno; Milford, Michael; Tuyen, Nguyen Tan Viet; Ramchurn, Sarvapali D.; Ehsan, Shoaib
On motion blur and deblurring in visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 10, no. 5, pp. 4746–4753, 2025.
Abstract | Links | BibTeX | Tags: data sets for robotic vision, Localization, Vision-Based Navigation
@article{soton507348,
title = {On motion blur and deblurring in visual place recognition},
author = {Timur Ismagilov and Bruno Ferrarini and Michael Milford and Nguyen Tan Viet Tuyen and Sarvapali D. Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/507348/},
year = {2025},
date = {2025-03-01},
journal = {IEEE Robotics and Automation Letters},
volume = {10},
number = {5},
pages = {4746–4753},
abstract = {ensuremath<pensuremath>Visual Place Recognition (VPR) in mobile robotics enables robots to localize themselves by recognizing previously visited locations using visual data. While the reliability of VPR methods has been extensively studied under conditions such as changes in illumination, season, weather and viewpoint, the impact of motion blur is relatively unexplored despite its relevance not only in rapid motion scenarios but also in low-light conditions where longer exposure times are necessary. Similarly, the role of image deblurring in enhancing VPR performance under motion blur has received limited attention so far. This letter bridges these gaps by introducing a new benchmark designed to evaluate VPR performance under the influence of motion blur and image deblurring. The benchmark includes three datasets that encompass a wide range of motion blur intensities, providing a comprehensive platform for analysis. Experimental results with several well-established VPR and image deblurring methods provide new insights into the effects of motion blur and the potential improvements achieved through deblurring. Building on these findings, the letter proposes adaptive deblurring strategies for VPR, designed to effectively manage motion blur in dynamic, real-world scenarios.ensuremath</pensuremath>},
keywords = {data sets for robotic vision, Localization, Vision-Based Navigation},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Visual Place Recognition (VPR) in mobile robotics enables robots to localize themselves by recognizing previously visited locations using visual data. While the reliability of VPR methods has been extensively studied under conditions such as changes in illumination, season, weather and viewpoint, the impact of motion blur is relatively unexplored despite its relevance not only in rapid motion scenarios but also in low-light conditions where longer exposure times are necessary. Similarly, the role of image deblurring in enhancing VPR performance under motion blur has received limited attention so far. This letter bridges these gaps by introducing a new benchmark designed to evaluate VPR performance under the influence of motion blur and image deblurring. The benchmark includes three datasets that encompass a wide range of motion blur intensities, providing a comprehensive platform for analysis. Experimental results with several well-established VPR and image deblurring methods provide new insights into the effects of motion blur and the potential improvements achieved through deblurring. Building on these findings, the letter proposes adaptive deblurring strategies for VPR, designed to effectively manage motion blur in dynamic, real-world scenarios.ensuremath</pensuremath>
Grainge, Oliver Edward; Milford, Michael J.; Bodala, Indu; Ramchurn, Sarvapali D.; Ehsan, Shoaib
Structured pruning for efficient visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 10, no. 2, pp. 2024–2031, 2025.
Abstract | Links | BibTeX | Tags: accuracy, computational modelling, convolutional neural network, feature extraction, Localization, memory management, Real-time systems, representation learning, robustness, Vision-Based Navigation, visual place recognition (VPR), visualization
@article{soton502843,
title = {Structured pruning for efficient visual place recognition},
author = {Oliver Edward Grainge and Michael J. Milford and Indu Bodala and Sarvapali D. Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/502843/},
year = {2025},
date = {2025-02-01},
journal = {IEEE Robotics and Automation Letters},
volume = {10},
number = {2},
pages = {2024–2031},
abstract = {Visual Place Recognition (VPR) is fundamental for the global re-localization of robots and devices, enabling them to recognize previously visited locations based on visual inputs. This capability is crucial for maintaining accurate mapping and localization over large areas. Given that VPR methods need to operate in real-time on embedded systems, it is critical to optimize these systems for minimal resource consumption. While the most efficient VPR approaches employ standard convolutional backbones with fixed descriptor dimensions, these often lead to redundancy in the embedding space as well as in the network architecture. Our work introduces a novel structured pruning method, to not only streamline common VPR architectures but also to strategically remove redundancies within the feature embedding space. This dual focus significantly enhances the efficiency of the system, reducing both map and model memory requirements and decreasing feature extraction and retrieval latencies. Our approach has reduced memory usage and latency by 21% and 16%, respectively, across models, while minimally impacting recall@1 accuracy by less than 1%. This significant improvement enhances real-time applications on edge devices with negligible accuracy loss.},
keywords = {accuracy, computational modelling, convolutional neural network, feature extraction, Localization, memory management, Real-time systems, representation learning, robustness, Vision-Based Navigation, visual place recognition (VPR), visualization},
pubstate = {published},
tppubtype = {article}
}
Visual Place Recognition (VPR) is fundamental for the global re-localization of robots and devices, enabling them to recognize previously visited locations based on visual inputs. This capability is crucial for maintaining accurate mapping and localization over large areas. Given that VPR methods need to operate in real-time on embedded systems, it is critical to optimize these systems for minimal resource consumption. While the most efficient VPR approaches employ standard convolutional backbones with fixed descriptor dimensions, these often lead to redundancy in the embedding space as well as in the network architecture. Our work introduces a novel structured pruning method, to not only streamline common VPR architectures but also to strategically remove redundancies within the feature embedding space. This dual focus significantly enhances the efficiency of the system, reducing both map and model memory requirements and decreasing feature extraction and retrieval latencies. Our approach has reduced memory usage and latency by 21% and 16%, respectively, across models, while minimally impacting recall@1 accuracy by less than 1%. This significant improvement enhances real-time applications on edge devices with negligible accuracy loss.
Kelly, Thomas Graham
Self-organised communication-aware control structures for robot swarms PhD Thesis
University of Southampton, 2025.
Abstract | Links | BibTeX | Tags:
@phdthesis{soton502200,
title = {Self-organised communication-aware control structures for robot swarms},
author = {Thomas Graham Kelly},
url = {https://eprints.soton.ac.uk/502200/},
year = {2025},
date = {2025-01-01},
publisher = {University of Southampton},
school = {University of Southampton},
abstract = {Robotic swarms are complex systems that rely on local communication between individual members of the swarm to spread information about their state and the environment. This information informs decisions and aids in tasks such as exploration and mapping. When communications break down between members of a swarm, it can become difficult to maintain accurate and up-to-date information about the state of the swarm and the environment. This problem is pertinent when humans are involved and may act as operators or teammates of the swarm. Here it is vital that the swarm can coordinate to distil and disseminate the vast amounts of information collected to the humans and throughout the swarm effectively, to maintain situational awareness.ensuremath<br/ensuremath>ensuremath<br/ensuremath>The collective decision-making of a swarm is one aspect that relies heavily on the ability to share information and observations to reach a swarm-wide consensus. This thesis investigates how communication constraints affect the swarm?s ability to reach a consensus and implement a communication-aware coordination strategy to mitigate these effects. We propose the communication-constrained collective decision-making problem and compare the performance of several collective decision-making strategies, enhanced with our coordination algorithm. We find that using such an approach improves the speed of a swarm to reach a consensus.ensuremath<br/ensuremath>ensuremath<br/ensuremath>Following on from this work, we examine a hybrid swarm system in a communication-limited environment. While swarms are traditionally considered decentralised systems, recent approaches have integrated decentralised and centralised control into a swarm system. We study the trade-offs in performance and communication in a hybrid system that can vary its control structure. We find that a higher level of centralisation does not guarantee higher performance and study how communication with a human operator is affected by the control structure. This work is extended to assess the feasibility of enabling a swarm system to learn the optimal control structure on the fly, according to mission requirements.},
keywords = {},
pubstate = {published},
tppubtype = {phdthesis}
}
Robotic swarms are complex systems that rely on local communication between individual members of the swarm to spread information about their state and the environment. This information informs decisions and aids in tasks such as exploration and mapping. When communications break down between members of a swarm, it can become difficult to maintain accurate and up-to-date information about the state of the swarm and the environment. This problem is pertinent when humans are involved and may act as operators or teammates of the swarm. Here it is vital that the swarm can coordinate to distil and disseminate the vast amounts of information collected to the humans and throughout the swarm effectively, to maintain situational awareness.ensuremath<br/ensuremath>ensuremath<br/ensuremath>The collective decision-making of a swarm is one aspect that relies heavily on the ability to share information and observations to reach a swarm-wide consensus. This thesis investigates how communication constraints affect the swarm?s ability to reach a consensus and implement a communication-aware coordination strategy to mitigate these effects. We propose the communication-constrained collective decision-making problem and compare the performance of several collective decision-making strategies, enhanced with our coordination algorithm. We find that using such an approach improves the speed of a swarm to reach a consensus.ensuremath<br/ensuremath>ensuremath<br/ensuremath>Following on from this work, we examine a hybrid swarm system in a communication-limited environment. While swarms are traditionally considered decentralised systems, recent approaches have integrated decentralised and centralised control into a swarm system. We study the trade-offs in performance and communication in a hybrid system that can vary its control structure. We find that a higher level of centralisation does not guarantee higher performance and study how communication with a human operator is affected by the control structure. This work is extended to assess the feasibility of enabling a swarm system to learn the optimal control structure on the fly, according to mission requirements.
Kiden, Sarah; Parisio, Isabela; Neff, Gina; Ramchurn, Gopal
AI management essentials (AIME) consultation response Technical Report
no. 10.5258/SOTON/PP0116, 2025.
Abstract | Links | BibTeX | Tags:
@techreport{soton501075,
title = {AI management essentials (AIME) consultation response},
author = {Sarah Kiden and Isabela Parisio and Gina Neff and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/501075/},
year = {2025},
date = {2025-01-01},
number = {10.5258/SOTON/PP0116},
publisher = {University of Southampton},
abstract = {We are submitting this response to the Department for Science, Innovation and Technology (DSIT) consultation on the new AI Management Essentials (AIME) tool1 on behalf of Responsible AI UK (RAi UK), an open and multidisciplinary network that brings together researchers from across the four nations of the UK to understand how we should shape the development of AI to benefit people, communities and society. To arrive at this response, we sent out a call to Principal Investigators and Co-Investigators of RAi UK funded projects2 to contribute to this consultation. What follows is a synthesis of the responses we received from our research community. Overall, RAi UK sees the AIME tool as a valuable first step for Small to Medium Sized Enterprises (SMEs) and Startups to adopt towards implementing robust and responsible AI governance practices. Our contributions below aim to improve the tool?s usability and impact.},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
We are submitting this response to the Department for Science, Innovation and Technology (DSIT) consultation on the new AI Management Essentials (AIME) tool1 on behalf of Responsible AI UK (RAi UK), an open and multidisciplinary network that brings together researchers from across the four nations of the UK to understand how we should shape the development of AI to benefit people, communities and society. To arrive at this response, we sent out a call to Principal Investigators and Co-Investigators of RAi UK funded projects2 to contribute to this consultation. What follows is a synthesis of the responses we received from our research community. Overall, RAi UK sees the AIME tool as a valuable first step for Small to Medium Sized Enterprises (SMEs) and Startups to adopt towards implementing robust and responsible AI governance practices. Our contributions below aim to improve the tool?s usability and impact.
Kiden, Sarah; Yazdanpanah, Vahid; Mestre, Rafael; Iusmen, Ingi; Atkinson, Joe; Parisio, Isabela; Ramchurn, Gopal; Stein, Sebastian
Human Rights and the Regulation of AI Written Evidence Technical Report
no. 10.5258/SOTON/PP0152, 2025.
@techreport{soton506667,
title = {Human Rights and the Regulation of AI Written Evidence},
author = {Sarah Kiden and Vahid Yazdanpanah and Rafael Mestre and Ingi Iusmen and Joe Atkinson and Isabela Parisio and Gopal Ramchurn and Sebastian Stein},
url = {https://eprints.soton.ac.uk/506667/},
year = {2025},
date = {2025-01-01},
number = {10.5258/SOTON/PP0152},
publisher = {University of Southampton},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
2024
Thavanesan, Navamayooran; Farahi, Arya; Parfitt, Charlotte; Belkhatir, Zehor; Azim, Tayyaba; Vallejos, Elvira Perez; Walters, Zoë; Ramchurn, Sarvapali; Underwood, Timothy J.; Vigneswaran, Ganesh
Insights from explainable AI in oesophageal cancer team decisions Journal Article
In: Computers in Biology and Medicine, vol. 180, 2024, (For the purpose of open access, the authors have applied a Creative Commons attribution license (CC-BY) to any Author Accepted Manuscript version arising from this submission.).
Abstract | Links | BibTeX | Tags: Decision-making, machine learning, Multidisciplinary teams, Oesophageal cancer
@article{soton493238,
title = {Insights from explainable AI in oesophageal cancer team decisions},
author = {Navamayooran Thavanesan and Arya Farahi and Charlotte Parfitt and Zehor Belkhatir and Tayyaba Azim and Elvira Perez Vallejos and Zoë Walters and Sarvapali Ramchurn and Timothy J. Underwood and Ganesh Vigneswaran},
url = {https://eprints.soton.ac.uk/493238/},
year = {2024},
date = {2024-08-01},
journal = {Computers in Biology and Medicine},
volume = {180},
abstract = {ensuremath<pensuremath>Background: clinician-led quality control into oncological decision-making is crucial for optimising patient care. Explainable artificial intelligence (XAI) techniques provide data-driven approaches to unravel how clinical variables influence this decision-making. We applied global XAI techniques to examine the impact of key clinical decision-drivers when mapped by a machine learning (ML) model, on the likelihood of receiving different oesophageal cancer (OC) treatment modalities by the multidisciplinary team (MDT).ensuremath</pensuremath>ensuremath<pensuremath>Methods: retrospective analysis of 893 OC patients managed between 2010 and 2022 at our tertiary unit, used a random forests (RF) classifier to predict four possible treatment pathways as determined by the MDT: neoadjuvant chemotherapy followed by surgery (NACT + S), neoadjuvant chemoradiotherapy followed by surgery (NACRT + S), surgery-alone, and palliative management. Variable importance and partial dependence (PD) analyses then examined the influence of targeted high-ranking clinical variables within the ML model on treatment decisions as a surrogate model of the MDT decision-making dynamic.�ensuremath</pensuremath>ensuremath<pensuremath>Results: amongst guideline-variables known to determine treatments, such as Tumour-Node-Metastasis (TNM) staging, age also proved highly important to the RF model (16.1 % of total importance) on variable importance analysis. PD subsequently revealed that predicted probabilities for all treatment modalities change significantly after 75 years (p < 0.001). Likelihood of surgery-alone and palliative therapies increased for patients aged 75?85yrs but lowered for NACT/NACRT. Performance status divided patients into two clusters which influenced all predicted outcomes in conjunction with age.�ensuremath</pensuremath>ensuremath<pensuremath>Conclusion: XAI techniques delineate the relationship between clinical factors and OC treatment decisions. These techniques identify advanced age as heavily influencing decisions based on our model with a greater role in patients with specific tumour characteristics. This study methodology provides the means for exploring conscious/subconscious bias and interrogating inconsistencies in team-based decision-making within the era of AI-driven decision support.ensuremath</pensuremath>},
note = {For the purpose of open access, the authors have applied a Creative Commons attribution license (CC-BY) to any Author Accepted Manuscript version arising from this submission.},
keywords = {Decision-making, machine learning, Multidisciplinary teams, Oesophageal cancer},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Background: clinician-led quality control into oncological decision-making is crucial for optimising patient care. Explainable artificial intelligence (XAI) techniques provide data-driven approaches to unravel how clinical variables influence this decision-making. We applied global XAI techniques to examine the impact of key clinical decision-drivers when mapped by a machine learning (ML) model, on the likelihood of receiving different oesophageal cancer (OC) treatment modalities by the multidisciplinary team (MDT).ensuremath</pensuremath>ensuremath<pensuremath>Methods: retrospective analysis of 893 OC patients managed between 2010 and 2022 at our tertiary unit, used a random forests (RF) classifier to predict four possible treatment pathways as determined by the MDT: neoadjuvant chemotherapy followed by surgery (NACT + S), neoadjuvant chemoradiotherapy followed by surgery (NACRT + S), surgery-alone, and palliative management. Variable importance and partial dependence (PD) analyses then examined the influence of targeted high-ranking clinical variables within the ML model on treatment decisions as a surrogate model of the MDT decision-making dynamic.�ensuremath</pensuremath>ensuremath<pensuremath>Results: amongst guideline-variables known to determine treatments, such as Tumour-Node-Metastasis (TNM) staging, age also proved highly important to the RF model (16.1 % of total importance) on variable importance analysis. PD subsequently revealed that predicted probabilities for all treatment modalities change significantly after 75 years (p < 0.001). Likelihood of surgery-alone and palliative therapies increased for patients aged 75?85yrs but lowered for NACT/NACRT. Performance status divided patients into two clusters which influenced all predicted outcomes in conjunction with age.�ensuremath</pensuremath>ensuremath<pensuremath>Conclusion: XAI techniques delineate the relationship between clinical factors and OC treatment decisions. These techniques identify advanced age as heavily influencing decisions based on our model with a greater role in patients with specific tumour characteristics. This study methodology provides the means for exploring conscious/subconscious bias and interrogating inconsistencies in team-based decision-making within the era of AI-driven decision support.ensuremath</pensuremath>
Naiseh, Mohammad; Webb, Catherine; Underwood, Tim; Ramchurn, Gopal; Walters, Zoe; Thavanesan, Navamayooran; Vigneswaran, Ganesh
XAI for group-AI interaction: towards collaborative and inclusive explanations Proceedings Article
In: Longo, Luca; Liu, Weiru; Montavon, Gregoire (Ed.): Joint Proceedings of the xAI 2024 Late-breaking Work, Demos and Doctoral Consortium co-located with the 2nd World Conference on eXplainable Artificial Intelligence (xAI 2024), pp. 249–256, CEUR Workshop Proceedings, 2024.
Abstract | Links | BibTeX | Tags: Explainable AI, Group-AI Interaction, Interaction Design
@inproceedings{soton497829,
title = {XAI for group-AI interaction: towards collaborative and inclusive explanations},
author = {Mohammad Naiseh and Catherine Webb and Tim Underwood and Gopal Ramchurn and Zoe Walters and Navamayooran Thavanesan and Ganesh Vigneswaran},
editor = {Luca Longo and Weiru Liu and Gregoire Montavon},
url = {https://eprints.soton.ac.uk/497829/},
year = {2024},
date = {2024-07-01},
booktitle = {Joint Proceedings of the xAI 2024 Late-breaking Work, Demos and Doctoral Consortium co-located with the 2nd World Conference on eXplainable Artificial Intelligence (xAI 2024)},
volume = {3793},
pages = {249–256},
publisher = {CEUR Workshop Proceedings},
abstract = {ensuremath<pensuremath>The increasing integration of Machine Learning (ML) into decision-making across various sectors has raised concerns about ethics, legality, explainability, and safety, highlighting the necessity of human oversight. In response, eXplainable AI (XAI) has emerged as a means to enhance transparency by providing insights into ML model decisions and offering humans an understanding of the underlying logic. Despite its potential, existing XAI models often lack practical usability and fail to improve human-AI performance, as they may introduce issues such as overreliance. This underscores the need for further research in Human-Centered XAI to improve the usability of current XAI methods. Notably, much of the current research focuses on one-to-one interactions between the XAI and individual decision-makers, overlooking the dynamics of many-to-one relationships in real-world scenarios where groups of humans collaborate using XAI in collective decision-making. In this late-breaking work, we draw upon current work in Human-Centered XAI research and discuss how XAI design could be transitioned to group-AI interaction. We discuss four potential challenges in the transition of XAI from human-AI interaction to group-AI interaction. This paper contributes to advancing the field of Human-Centered XAI and facilitates the discussion on group-XAI interaction, calling for further research in this area.ensuremath</pensuremath>},
keywords = {Explainable AI, Group-AI Interaction, Interaction Design},
pubstate = {published},
tppubtype = {inproceedings}
}
ensuremath<pensuremath>The increasing integration of Machine Learning (ML) into decision-making across various sectors has raised concerns about ethics, legality, explainability, and safety, highlighting the necessity of human oversight. In response, eXplainable AI (XAI) has emerged as a means to enhance transparency by providing insights into ML model decisions and offering humans an understanding of the underlying logic. Despite its potential, existing XAI models often lack practical usability and fail to improve human-AI performance, as they may introduce issues such as overreliance. This underscores the need for further research in Human-Centered XAI to improve the usability of current XAI methods. Notably, much of the current research focuses on one-to-one interactions between the XAI and individual decision-makers, overlooking the dynamics of many-to-one relationships in real-world scenarios where groups of humans collaborate using XAI in collective decision-making. In this late-breaking work, we draw upon current work in Human-Centered XAI research and discuss how XAI design could be transitioned to group-AI interaction. We discuss four potential challenges in the transition of XAI from human-AI interaction to group-AI interaction. This paper contributes to advancing the field of Human-Centered XAI and facilitates the discussion on group-XAI interaction, calling for further research in this area.ensuremath</pensuremath>
Early, Joseph Arthur
Interpretable multiple instance learning PhD Thesis
University of Southampton, 2024.
Abstract | Links | BibTeX | Tags:
@phdthesis{soton490767,
title = {Interpretable multiple instance learning},
author = {Joseph Arthur Early},
url = {https://eprints.soton.ac.uk/490767/},
year = {2024},
date = {2024-06-01},
publisher = {University of Southampton},
school = {University of Southampton},
abstract = {With the rising use of Artificial Intelligence (AI) and Machine Learning (ML) methods, there comes an increasing need to understand how automated systems make decisions. Interpretable ML provides insight into the underlying reasoning behind AI and ML models while not stifling their predictive performance. Doing so is important for many reasons, such as facilitating trust, increasing transparency, and providing improved collaboration and control through a better understanding of automated decision-making. Interpretability is very relevant across many ML paradigms and application domains. Multiple Instance Learning (MIL) is an ML paradigm where data are grouped into bags of instances, and only the bags are labelled (rather than each instance). This is beneficial in alleviating expensive labelling procedures and can be used to exploit the underlying structure of data. This thesis investigates how interpretability can be achieved within MIL. It begins with a formalisation of interpretable MIL, and then proposes a suite of model-agnostic post-hoc methods. This work is then extended to the specific application domain of high-resolution satellite imagery, using novel inherently interpretable MIL approaches that operate at multiple resolutions. Following on from work in the vision domain, new methods for interpretable MIL are developed for sequential data. First, it is explored in the domain of Reward Modelling (RM) for Reinforcement Learning (RL), demonstrating that interpretable MIL can be used to not only understand a model but also improve its predictive performance. This is mirrored in the application of interpretable MIL to Time Series Classification (TSC), where it is integrated into state-of-the-art methods and is able to improve both their interpretability and predictive performance. The integration into existing models to provide inherent interpretability means these benefits are delivered with little additional computational cost. ensuremath<br/ensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {phdthesis}
}
With the rising use of Artificial Intelligence (AI) and Machine Learning (ML) methods, there comes an increasing need to understand how automated systems make decisions. Interpretable ML provides insight into the underlying reasoning behind AI and ML models while not stifling their predictive performance. Doing so is important for many reasons, such as facilitating trust, increasing transparency, and providing improved collaboration and control through a better understanding of automated decision-making. Interpretability is very relevant across many ML paradigms and application domains. Multiple Instance Learning (MIL) is an ML paradigm where data are grouped into bags of instances, and only the bags are labelled (rather than each instance). This is beneficial in alleviating expensive labelling procedures and can be used to exploit the underlying structure of data. This thesis investigates how interpretability can be achieved within MIL. It begins with a formalisation of interpretable MIL, and then proposes a suite of model-agnostic post-hoc methods. This work is then extended to the specific application domain of high-resolution satellite imagery, using novel inherently interpretable MIL approaches that operate at multiple resolutions. Following on from work in the vision domain, new methods for interpretable MIL are developed for sequential data. First, it is explored in the domain of Reward Modelling (RM) for Reinforcement Learning (RL), demonstrating that interpretable MIL can be used to not only understand a model but also improve its predictive performance. This is mirrored in the application of interpretable MIL to Time Series Classification (TSC), where it is integrated into state-of-the-art methods and is able to improve both their interpretability and predictive performance. The integration into existing models to provide inherent interpretability means these benefits are delivered with little additional computational cost. ensuremath<br/ensuremath>
Kiden, Sarah; Stahl, Bernd; Townsend, Beverley; Maple, Carsten; Vincent, Charles; Sampson, Fraser; Gilbert, Geoff; Smith, Helen; Deshmukh, Jayati; Ross, Jen; Williams, Jennifer; Rincon, Jesus Martinez; Lisinska, Justyna; O?Shea, Karen; Abreu, Márjory Da Costa; Bencomo, Nelly; Deb, Oishi; Winter, Peter; Li, Phoebe; Torr, Philip; Lau, Pin Lean; Iniesta, Raquel; Ramchurn, Gopal; Stein, Sebastian; Yazdanpanah, Vahid
Responsible AI governance: A response to UN interim report on governing AI for humanity Technical Report
no. 10.5258/SOTON/PP0057, 2024.
@techreport{soton488908,
title = {Responsible AI governance: A response to UN interim report on governing AI for humanity},
author = {Sarah Kiden and Bernd Stahl and Beverley Townsend and Carsten Maple and Charles Vincent and Fraser Sampson and Geoff Gilbert and Helen Smith and Jayati Deshmukh and Jen Ross and Jennifer Williams and Jesus Martinez Rincon and Justyna Lisinska and Karen O?Shea and Márjory Da Costa Abreu and Nelly Bencomo and Oishi Deb and Peter Winter and Phoebe Li and Philip Torr and Pin Lean Lau and Raquel Iniesta and Gopal Ramchurn and Sebastian Stein and Vahid Yazdanpanah},
url = {https://eprints.soton.ac.uk/488908/},
year = {2024},
date = {2024-03-01},
number = {10.5258/SOTON/PP0057},
publisher = {Public Policy, University of Southampton},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
Abioye, Ayodeji O.; Hunt, William; Gu, Yue; Schneiders, Eike; Naiseh, Mohammad; Fischer, Joel E.; Ramchurn, Sarvapali D.; Soorati, Mohammad D.; Archibald, Blair; Sevegnani, Michele
The effect of predictive formal modelling at runtime on performance in human-swarm interaction Proceedings Article
In: HRI '24: Companion of the 2024 ACM/IEEE International Conference on Human-Robot Interaction, pp. 172?176, Association for Computing Machinery, 2024, (Publisher Copyright: © 2024 Copyright held by the owner/author(s)).
Abstract | Links | BibTeX | Tags: Human-Robot Interaction (HRI), Human-Swarm Interaction (HSI), Predictive Formal Modelling (PFM), Task Performance
@inproceedings{soton488273,
title = {The effect of predictive formal modelling at runtime on performance in human-swarm interaction},
author = {Ayodeji O. Abioye and William Hunt and Yue Gu and Eike Schneiders and Mohammad Naiseh and Joel E. Fischer and Sarvapali D. Ramchurn and Mohammad D. Soorati and Blair Archibald and Michele Sevegnani},
url = {https://eprints.soton.ac.uk/488273/},
year = {2024},
date = {2024-03-01},
booktitle = {HRI '24: Companion of the 2024 ACM/IEEE International Conference on Human-Robot Interaction},
pages = {172?176},
publisher = {Association for Computing Machinery},
abstract = {Formal Modelling is often used as part of the design and testing process of software development to ensure that components operate within suitable bounds even in unexpected circumstances. In this paper, we use predictive formal modelling (PFM) at runtime in a human-swarm mission and show that this integration can be used to improve the performance of human-swarm teams. We recruited 60 participants to operate a simulated aerial swarm to deliver parcels to target locations. In the PFM condition, operators were informed of the estimated completion times given the number of drones deployed, whereas, in the No-PFM condition, operators did not have this information. The operators could control the mission by adding or removing drones from the mission and thereby, increasing or decreasing the overall mission cost. The evaluation of human-swarm performance relied on four metrics: the task completion time, the number of agents, the number of completed tasks, and the cost per task. Our results show that PFM modelling at runtime improves mission performance without significantly affecting the operator's workload or the system's usability.},
note = {Publisher Copyright:
© 2024 Copyright held by the owner/author(s)},
keywords = {Human-Robot Interaction (HRI), Human-Swarm Interaction (HSI), Predictive Formal Modelling (PFM), Task Performance},
pubstate = {published},
tppubtype = {inproceedings}
}
Formal Modelling is often used as part of the design and testing process of software development to ensure that components operate within suitable bounds even in unexpected circumstances. In this paper, we use predictive formal modelling (PFM) at runtime in a human-swarm mission and show that this integration can be used to improve the performance of human-swarm teams. We recruited 60 participants to operate a simulated aerial swarm to deliver parcels to target locations. In the PFM condition, operators were informed of the estimated completion times given the number of drones deployed, whereas, in the No-PFM condition, operators did not have this information. The operators could control the mission by adding or removing drones from the mission and thereby, increasing or decreasing the overall mission cost. The evaluation of human-swarm performance relied on four metrics: the task completion time, the number of agents, the number of completed tasks, and the cost per task. Our results show that PFM modelling at runtime improves mission performance without significantly affecting the operator's workload or the system's usability.
Soorati, Mohammad D.; Naiseh, Mohammad; Hunt, William; Parnell, Katie; Clark, Jediah; Ramchurn, Sarvapali D.
Enabling trustworthiness in human-swarm systems through a digital twin Book Section
In: Dasgupta, Prithviraj; Llinas, James; Gillespie, Tony; Fouse, Scott; Lawless, William; Mittu, Ranjeev; Sofge, Donlad (Ed.): Putting AI in the Critical Loop: Assured Trust and Autonomy in Human-Machine Teams, pp. 93–125, Academic Press, 2024, (Publisher Copyright: © 2024 Elsevier Inc. All rights reserved.).
Abstract | Links | BibTeX | Tags: Digital twin, Explainability, Human-swarm interaction, Trustworthy Autonomous Systems, User-centered design
@incollection{soton491769,
title = {Enabling trustworthiness in human-swarm systems through a digital twin},
author = {Mohammad D. Soorati and Mohammad Naiseh and William Hunt and Katie Parnell and Jediah Clark and Sarvapali D. Ramchurn},
editor = {Prithviraj Dasgupta and James Llinas and Tony Gillespie and Scott Fouse and William Lawless and Ranjeev Mittu and Donlad Sofge},
url = {https://eprints.soton.ac.uk/491769/},
year = {2024},
date = {2024-02-01},
booktitle = {Putting AI in the Critical Loop: Assured Trust and Autonomy in Human-Machine Teams},
pages = {93–125},
publisher = {Academic Press},
abstract = {Robot swarms are highly dynamic systems that exhibit fault-tolerant behavior in accomplishing given tasks. Applications of swarm robotics are very limited due to the lack of complex decision-making capability. Real-world applications are only possible if we use human supervision to monitor and control the behavior of the swarm. Ensuring that human operators can trust the swarm system is one of the key challenges in human-swarm systems. This chapter presents a digital twin for trustworthy human-swarm teaming. The first element in designing such a simulation platform is to understand the trust requirements to label a human-swarm system as trustworthy. In order to outline the key trust requirements, we interviewed a group of experienced uncrewed aerial vehicle (UAV) operators and collated their suggestions for building and repairing trusts in single and multiple UAV systems. We then performed a survey to gather swarm experts? points of view on creating a taxonomy for explainability in human-swarm systems. This chapter presents a digital twin platform that implements a disaster management use case and has the capacity to meet the extracted trust and explainability requirements.},
note = {Publisher Copyright:
© 2024 Elsevier Inc. All rights reserved.},
keywords = {Digital twin, Explainability, Human-swarm interaction, Trustworthy Autonomous Systems, User-centered design},
pubstate = {published},
tppubtype = {incollection}
}
Robot swarms are highly dynamic systems that exhibit fault-tolerant behavior in accomplishing given tasks. Applications of swarm robotics are very limited due to the lack of complex decision-making capability. Real-world applications are only possible if we use human supervision to monitor and control the behavior of the swarm. Ensuring that human operators can trust the swarm system is one of the key challenges in human-swarm systems. This chapter presents a digital twin for trustworthy human-swarm teaming. The first element in designing such a simulation platform is to understand the trust requirements to label a human-swarm system as trustworthy. In order to outline the key trust requirements, we interviewed a group of experienced uncrewed aerial vehicle (UAV) operators and collated their suggestions for building and repairing trusts in single and multiple UAV systems. We then performed a survey to gather swarm experts? points of view on creating a taxonomy for explainability in human-swarm systems. This chapter presents a digital twin platform that implements a disaster management use case and has the capacity to meet the extracted trust and explainability requirements.
Thavanesan, Navamayooran; Parfitt, Charlotte; Bodala, Indu; Walters, Zoë; Ramchurn, Sarvapali; Underwood, Timothy; Vigneswaran, Ganesh
Machine learning models for curative and palliative oesophageal cancer treatment pathway prediction Miscellaneous
2024.
Abstract | Links | BibTeX | Tags:
@misc{soton497828,
title = {Machine learning models for curative and palliative oesophageal cancer treatment pathway prediction},
author = {Navamayooran Thavanesan and Charlotte Parfitt and Indu Bodala and Zoë Walters and Sarvapali Ramchurn and Timothy Underwood and Ganesh Vigneswaran},
url = {https://eprints.soton.ac.uk/497828/},
year = {2024},
date = {2024-01-01},
journal = {European Journal of Surgical Oncology},
volume = {50},
number = {1},
abstract = {Introduction: Oesophageal Cancer Multidisciplinary Teams (OC MDTs) operate under significant caseload pressures. This risks variability of decision-making which may influence patient outcomes. Machine Learning (ML) offers the ability to streamline and standardise decision-making by learning from historic treatment decisions to prediction treatment for new patients. We present internally validated ML models designed to predict OC MDT treatment decisions for curative and palliative OC patients.ensuremath<br/ensuremath>ensuremath<br/ensuremath>Methods: four ML algorithms (multinomial logistic regression (MLR), random forests (RF), extreme gradient boost (XGB) and decision tree (DT)) were trained using nested cross-validation on a cohort of 938 OC cases from a single tertiary unit over a 12-year period. The models classified predicted treatments into one of: Surgery (S), Neoadjuvant Chemotherapy (NACT) + S, Neoadjuvant Chemoradiotherapy (NACRT) + S, Endoscopic or Palliative treatment. Performance was assessed on Area Under the Curve (AUC).ensuremath<br/ensuremath>ensuremath<br/ensuremath>Results: across algorithms, all models performed strongly with mean AUC for Surgery = 0.849$±$0.026, NACT +S = 0.884$±$0.008, NACRT +S = 0.834$±$0.035},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Introduction: Oesophageal Cancer Multidisciplinary Teams (OC MDTs) operate under significant caseload pressures. This risks variability of decision-making which may influence patient outcomes. Machine Learning (ML) offers the ability to streamline and standardise decision-making by learning from historic treatment decisions to prediction treatment for new patients. We present internally validated ML models designed to predict OC MDT treatment decisions for curative and palliative OC patients.ensuremath<br/ensuremath>ensuremath<br/ensuremath>Methods: four ML algorithms (multinomial logistic regression (MLR), random forests (RF), extreme gradient boost (XGB) and decision tree (DT)) were trained using nested cross-validation on a cohort of 938 OC cases from a single tertiary unit over a 12-year period. The models classified predicted treatments into one of: Surgery (S), Neoadjuvant Chemotherapy (NACT) + S, Neoadjuvant Chemoradiotherapy (NACRT) + S, Endoscopic or Palliative treatment. Performance was assessed on Area Under the Curve (AUC).ensuremath<br/ensuremath>ensuremath<br/ensuremath>Results: across algorithms, all models performed strongly with mean AUC for Surgery = 0.849$±$0.026, NACT +S = 0.884$±$0.008, NACRT +S = 0.834$±$0.035
Kelly, Thomas Graham; Soorati, Mohammad; Zauner, Klaus-Peter; Ramchurn, Gopal; Tarapore, Danesh
Trade-offs of dynamic control structure in human-swarm systems Proceedings Article
In: The International Symposium on Distributed Autonomous Robotic Systems (DARS) 2024, 2024.
Abstract | Links | BibTeX | Tags:
@inproceedings{soton492838,
title = {Trade-offs of dynamic control structure in human-swarm systems},
author = {Thomas Graham Kelly and Mohammad Soorati and Klaus-Peter Zauner and Gopal Ramchurn and Danesh Tarapore},
url = {https://eprints.soton.ac.uk/492838/},
year = {2024},
date = {2024-01-01},
booktitle = {The International Symposium on Distributed Autonomous Robotic Systems (DARS) 2024},
abstract = {Swarm robotics is a study of simple robots that exhibit complex behaviour only by interacting locally with other robots and their environment. The control in swarm robotics is mainly distributed whereas centralised control is widely used in other fields of robotics. Centralised and decentralised control strategies both pose a unique set of benefits and drawbacks for the control of multi-robot systems. While decentralised systems are more scalable and resilient, they are less efficient compared to the centralised systems and they lead to excessive data transmissions to the human operators causing cognitive overload. We examine the trade-offs of each of these approaches in a human-swarm system to perform an environmental monitoring task and propose a flexible hybrid approach, which combines elements of hierarchical and decentralised systems. We find that a flexible hybrid system can outperform a centralised system (in our environmental monitoring task by 19.2%) while reducing the number of messages sent to a human operator (here by 23.1%). We conclude that establishing centralisation for a system is not always optimal for performance and that utilising aspects of centralised and decentralised systems can keep the swarm from hindering its performance.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Swarm robotics is a study of simple robots that exhibit complex behaviour only by interacting locally with other robots and their environment. The control in swarm robotics is mainly distributed whereas centralised control is widely used in other fields of robotics. Centralised and decentralised control strategies both pose a unique set of benefits and drawbacks for the control of multi-robot systems. While decentralised systems are more scalable and resilient, they are less efficient compared to the centralised systems and they lead to excessive data transmissions to the human operators causing cognitive overload. We examine the trade-offs of each of these approaches in a human-swarm system to perform an environmental monitoring task and propose a flexible hybrid approach, which combines elements of hierarchical and decentralised systems. We find that a flexible hybrid system can outperform a centralised system (in our environmental monitoring task by 19.2%) while reducing the number of messages sent to a human operator (here by 23.1%). We conclude that establishing centralisation for a system is not always optimal for performance and that utilising aspects of centralised and decentralised systems can keep the swarm from hindering its performance.
2023
Early, Joseph; Deweese, Ying-Jung Chen; Evers, Christine; Ramchurn, Sarvapali
Extending scene-to-patch models: Multi-resolution multiple instance learning for Earth observation Journal Article
In: Environmental Data Science, vol. 2, pp. 18, 2023.
Abstract | Links | BibTeX | Tags:
@article{soton490766,
title = {Extending scene-to-patch models: Multi-resolution multiple instance learning for Earth observation},
author = {Joseph Early and Ying-Jung Chen Deweese and Christine Evers and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/490766/},
year = {2023},
date = {2023-12-01},
journal = {Environmental Data Science},
volume = {2},
pages = {18},
abstract = {Land cover classification (LCC) and natural disaster response (NDR) are important issues in climate change mitigation and adaptation. Existing approaches that use machine learning with Earth observation (EO) imaging data for LCC and NDR often rely on fully annotated and segmented datasets. Creating these datasets requires a large amount of effort, and a lack of suitable datasets has become an obstacle in scaling the use of machine learning for EO. In this study, we extend our prior work on Scene-to-Patch models: an alternative machine learning approach for EO that utilizes Multiple Instance Learning (MIL). As our approach only requires high-level scene labels, it enables much faster development of new datasets while still providing segmentation through patch-level predictions, ultimately increasing the accessibility of using machine learning for EO. We propose new multi-resolution MIL architectures that outperform single-resolution MIL models and non-MIL baselines on the DeepGlobe LCC and FloodNet NDR datasets. In addition, we conduct a thorough analysis of model performance and interpretability.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Land cover classification (LCC) and natural disaster response (NDR) are important issues in climate change mitigation and adaptation. Existing approaches that use machine learning with Earth observation (EO) imaging data for LCC and NDR often rely on fully annotated and segmented datasets. Creating these datasets requires a large amount of effort, and a lack of suitable datasets has become an obstacle in scaling the use of machine learning for EO. In this study, we extend our prior work on Scene-to-Patch models: an alternative machine learning approach for EO that utilizes Multiple Instance Learning (MIL). As our approach only requires high-level scene labels, it enables much faster development of new datasets while still providing segmentation through patch-level predictions, ultimately increasing the accessibility of using machine learning for EO. We propose new multi-resolution MIL architectures that outperform single-resolution MIL models and non-MIL baselines on the DeepGlobe LCC and FloodNet NDR datasets. In addition, we conduct a thorough analysis of model performance and interpretability.
Rigley, Eryn; Bentley, Caitlin; Krook, Joshua; Ramchurn, Gopal
Evaluating international AI skills policy: a systematic review of AI skills policy in seven countries Journal Article
In: Global Policy, 2023, (Funding Information: This research was supported via UKRI by the DCMS Science and Analysis R&D Programme. It was developed and produced according to UKRI's initial hypotheses and output requests. Any primary research, subsequent findings or recommendations do not represent Government views or policy and are produced according to academic ethics, quality assurance and independence.).
Abstract | Links | BibTeX | Tags:
@article{soton485727,
title = {Evaluating international AI skills policy: a systematic review of AI skills policy in seven countries},
author = {Eryn Rigley and Caitlin Bentley and Joshua Krook and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/485727/},
year = {2023},
date = {2023-12-01},
journal = {Global Policy},
abstract = {ensuremath<pensuremath>As artificial intelligence (AI) is having an increasingly disruptive impact across industries, companies continue to report having difficulty when recruiting for AI roles, while new graduates find it difficult to find employment, indicating a skills gap or skills misalignment. International approaches to AI skills programmes can offer a guide to future policy development of a skilled workforce, best placed to harness the economic opportunities that AI may support. The authors performed a systematic literature review on AI skills in government policies and documents from seven countries: Australia, Canada, China, Singapore, Sweden, the United Kingom and the United States. We found a divide between countries which emphasised a broader, nationwide approach to upskill and educate all citizens at different levels, namely the United States and Singapore and those countries which emphasised a narrower focus on educating a smaller group of experts with advanced AI knowledge and skills, namely China, Sweden and Canada. We found that the former, broader approaches tended to correlate with higher AI readiness and index scores than the narrower, expert-driven approach. Our findings indicate that, to match world-leading AI readiness, future AI skills policy should follow these broad, nationwide approaches to upskill and educate all citizens at different levels of AI expertise.ensuremath</pensuremath>},
note = {Funding Information:
This research was supported via UKRI by the DCMS Science and Analysis R&D Programme. It was developed and produced according to UKRI's initial hypotheses and output requests. Any primary research, subsequent findings or recommendations do not represent Government views or policy and are produced according to academic ethics, quality assurance and independence.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>As artificial intelligence (AI) is having an increasingly disruptive impact across industries, companies continue to report having difficulty when recruiting for AI roles, while new graduates find it difficult to find employment, indicating a skills gap or skills misalignment. International approaches to AI skills programmes can offer a guide to future policy development of a skilled workforce, best placed to harness the economic opportunities that AI may support. The authors performed a systematic literature review on AI skills in government policies and documents from seven countries: Australia, Canada, China, Singapore, Sweden, the United Kingom and the United States. We found a divide between countries which emphasised a broader, nationwide approach to upskill and educate all citizens at different levels, namely the United States and Singapore and those countries which emphasised a narrower focus on educating a smaller group of experts with advanced AI knowledge and skills, namely China, Sweden and Canada. We found that the former, broader approaches tended to correlate with higher AI readiness and index scores than the narrower, expert-driven approach. Our findings indicate that, to match world-leading AI readiness, future AI skills policy should follow these broad, nationwide approaches to upskill and educate all citizens at different levels of AI expertise.ensuremath</pensuremath>
Singh, Lokesh; Ramchurn, Gopal
The effect of automated agents on individual performance under induced stress Proceedings Article
In: Kalra, Jay (Ed.): Emerging Technologies in Healthcare and Medicine: Proceedings of the AHFE International Conference on Human Factors in Design, Engineering and Computing (AHFE 2023 Hawaii Edition), pp. 118–127, AHFE International, 2023.
Abstract | Links | BibTeX | Tags: Decision-making, Human-agent, Individual performance, Induced stress, Time pressure
@inproceedings{soton485655,
title = {The effect of automated agents on individual performance under induced stress},
author = {Lokesh Singh and Gopal Ramchurn},
editor = {Jay Kalra},
url = {https://eprints.soton.ac.uk/485655/},
year = {2023},
date = {2023-11-01},
booktitle = {Emerging Technologies in Healthcare and Medicine: Proceedings of the AHFE International Conference on Human Factors in Design, Engineering and Computing (AHFE 2023 Hawaii Edition)},
pages = {118–127},
publisher = {AHFE International},
abstract = {Induced stress is a phenomenon commonly experienced across different fields such as emergency services, healthcare, air traffic control, sports, and business - which necessitates the development of effective coping strategies and resilience for individuals or teams performing under pressure. This study aims to examine the effects of automated agents on individual performance during high-stress conditions. The design of these agents ensures they carry out identical tasks as participants based on predetermined frameworks. Participants underwent an experimentally designed task that aimed at inducing stress while measuring their performance amidst time pressure and auditory distraction. Results indicate that working with automated agents causes individuals to alter their approach by focusing narrowly on immediate concerns - making it challenging for them to consider several options or see broader contexts accurately. Regardless of ability level participants' performances were influenced by these automated agents. Future research will explore how these findings interact with physiological signals. This study highlights the importance of developing effective coping strategies and the potential impact of social factors on individual performance under induced stress.},
keywords = {Decision-making, Human-agent, Individual performance, Induced stress, Time pressure},
pubstate = {published},
tppubtype = {inproceedings}
}
Induced stress is a phenomenon commonly experienced across different fields such as emergency services, healthcare, air traffic control, sports, and business - which necessitates the development of effective coping strategies and resilience for individuals or teams performing under pressure. This study aims to examine the effects of automated agents on individual performance during high-stress conditions. The design of these agents ensures they carry out identical tasks as participants based on predetermined frameworks. Participants underwent an experimentally designed task that aimed at inducing stress while measuring their performance amidst time pressure and auditory distraction. Results indicate that working with automated agents causes individuals to alter their approach by focusing narrowly on immediate concerns - making it challenging for them to consider several options or see broader contexts accurately. Regardless of ability level participants' performances were influenced by these automated agents. Future research will explore how these findings interact with physiological signals. This study highlights the importance of developing effective coping strategies and the potential impact of social factors on individual performance under induced stress.
Thavanesan, Navamayooran; Bodala, Indu; Walters, Zoe; Ramchurn, Sarvapali; Underwood, Timothy J.; Vigneswaran, Ganesh
Machine learning to predict curative multidisciplinary team treatment decisions in oesophageal cancer Journal Article
In: European Journal of Surgical Oncology, vol. 49, no. 11, 2023, (Publisher Copyright: © 2023 The Author(s)).
Abstract | Links | BibTeX | Tags: Artificial Intelligence, machine learning, Oesophageal cancer multidisciplinary team
@article{soton479497b,
title = {Machine learning to predict curative multidisciplinary team treatment decisions in oesophageal cancer},
author = {Navamayooran Thavanesan and Indu Bodala and Zoe Walters and Sarvapali Ramchurn and Timothy J. Underwood and Ganesh Vigneswaran},
url = {https://eprints.soton.ac.uk/479497/},
year = {2023},
date = {2023-11-01},
journal = {European Journal of Surgical Oncology},
volume = {49},
number = {11},
abstract = {ensuremath<pensuremath>Background: Rising workflow pressures within the oesophageal cancer (OC) multidisciplinary team (MDT) can lead to variability in decision-making, and health inequality. Machine learning (ML) offers a potential automated data-driven approach to address inconsistency and standardize care. The aim of this experimental pilot study was to develop ML models able to predict curative OC MDT treatment decisions and determine the relative importance of underlying decision-critical variables. Methods: Retrospective complete-case analysis of oesophagectomy patients $±$ neoadjuvant chemotherapy (NACT) or chemoradiotherapy (NACRT) between 2010 and 2020. Established ML algorithms (Multinomial Logistic regression (MLR), Random Forests (RF), Extreme Gradient Boosting (XGB)) and Decision Tree (DT) were used to train models predicting OC MDT treatment decisions: surgery (S), NACT + S or NACRT + S. Performance metrics included Area Under the Curve (AUC), Accuracy, Kappa, LogLoss, F1 and Precision -Recall AUC. Variable importance was calculated for each model. Results: We identified 399 cases with a male-to-female ratio of 3.6:1 and median age of 66.1yrs (range 32?83). MLR outperformed RF, XGB and DT across performance metrics (mean AUC of 0.793 [$±$0.045] vs 0.757 [$±$0.068], 0.740 [$±$0.042], and 0.709 [$±$0.021] respectively). Variable importance analysis identified age as a major factor in the decision to offer surgery alone or NACT + S across models (p < 0.05). Conclusions: ML techniques can use limited feature-sets to predict curative UGI MDT treatment decisions. Explainable Artificial Intelligence methods provide insight into decision-critical variables, highlighting underlying subconscious biases in cancer care decision-making. Such models may allow prioritization of caseload, improve efficiency, and offer data-driven decision-assistance to MDTs in the future.ensuremath</pensuremath>},
note = {Publisher Copyright:
© 2023 The Author(s)},
keywords = {Artificial Intelligence, machine learning, Oesophageal cancer multidisciplinary team},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Background: Rising workflow pressures within the oesophageal cancer (OC) multidisciplinary team (MDT) can lead to variability in decision-making, and health inequality. Machine learning (ML) offers a potential automated data-driven approach to address inconsistency and standardize care. The aim of this experimental pilot study was to develop ML models able to predict curative OC MDT treatment decisions and determine the relative importance of underlying decision-critical variables. Methods: Retrospective complete-case analysis of oesophagectomy patients $±$ neoadjuvant chemotherapy (NACT) or chemoradiotherapy (NACRT) between 2010 and 2020. Established ML algorithms (Multinomial Logistic regression (MLR), Random Forests (RF), Extreme Gradient Boosting (XGB)) and Decision Tree (DT) were used to train models predicting OC MDT treatment decisions: surgery (S), NACT + S or NACRT + S. Performance metrics included Area Under the Curve (AUC), Accuracy, Kappa, LogLoss, F1 and Precision -Recall AUC. Variable importance was calculated for each model. Results: We identified 399 cases with a male-to-female ratio of 3.6:1 and median age of 66.1yrs (range 32?83). MLR outperformed RF, XGB and DT across performance metrics (mean AUC of 0.793 [$±$0.045] vs 0.757 [$±$0.068], 0.740 [$±$0.042], and 0.709 [$±$0.021] respectively). Variable importance analysis identified age as a major factor in the decision to offer surgery alone or NACT + S across models (p < 0.05). Conclusions: ML techniques can use limited feature-sets to predict curative UGI MDT treatment decisions. Explainable Artificial Intelligence methods provide insight into decision-critical variables, highlighting underlying subconscious biases in cancer care decision-making. Such models may allow prioritization of caseload, improve efficiency, and offer data-driven decision-assistance to MDTs in the future.ensuremath</pensuremath>
Krook, Joshua; Williams, Jennifer; Seabrooke, Tina; Schneiders, Eike; Blockx, Jan; Middleton, Stuart E; Ramchurn, Sarvapali
AI large language models inquiry: TASHub Response Miscellaneous
2023.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, chatgpt, european law, generative ai, Large Language Models, law and technology, technology policy
@misc{soton481740,
title = {AI large language models inquiry: TASHub Response},
author = {Joshua Krook and Jennifer Williams and Tina Seabrooke and Eike Schneiders and Jan Blockx and Stuart E Middleton and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/481740/},
year = {2023},
date = {2023-08-01},
publisher = {University of Southampton},
abstract = {Policy submission to the Consultation by Communications and Digital Committee, House of Lords, AI Large Language Models Inquiry.ensuremath<br/ensuremath>},
keywords = {Artificial Intelligence, chatgpt, european law, generative ai, Large Language Models, law and technology, technology policy},
pubstate = {published},
tppubtype = {misc}
}
Policy submission to the Consultation by Communications and Digital Committee, House of Lords, AI Large Language Models Inquiry.ensuremath<br/ensuremath>
Krook, Joshua; Williams, Jennifer; Seabrooke, Tina; Schneiders, Eike; Blockx, Jan; Middleton, Stuart E; Ramchurn, Sarvapali
AI large language models inquiry: TASHub response Miscellaneous
2023.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, chatgpt, european law, generative ai, Large Language Models, law and technology, technology policy
@misc{soton481740b,
title = {AI large language models inquiry: TASHub response},
author = {Joshua Krook and Jennifer Williams and Tina Seabrooke and Eike Schneiders and Jan Blockx and Stuart E Middleton and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/481740/},
year = {2023},
date = {2023-08-01},
publisher = {University of Southampton},
abstract = {Policy submission to the Consultation by Communications and Digital Committee, House of Lords, AI Large Language Models Inquiry.ensuremath<br/ensuremath>},
keywords = {Artificial Intelligence, chatgpt, european law, generative ai, Large Language Models, law and technology, technology policy},
pubstate = {published},
tppubtype = {misc}
}
Policy submission to the Consultation by Communications and Digital Committee, House of Lords, AI Large Language Models Inquiry.ensuremath<br/ensuremath>
Thavanesan, Navamayooran; Bodala, Indu; Walters, Zoe; Ramchurn, Sarvapali; Underwood, Timothy J; Vigneswaran, Ganesh
Machine learning to predict curative multidisciplinary team treatment decisions in oesophageal cancer Journal Article
In: European Journal of Surgical Oncology, 2023, (Publisher Copyright: copyright 2023 The Author(s)).
Abstract | Links | BibTeX | Tags: Artificial Intelligence, machine learning, Oesophageal cancer multidisciplinary team
@article{soton479497,
title = {Machine learning to predict curative multidisciplinary team treatment decisions in oesophageal cancer},
author = {Navamayooran Thavanesan and Indu Bodala and Zoe Walters and Sarvapali Ramchurn and Timothy J Underwood and Ganesh Vigneswaran},
url = {https://eprints.soton.ac.uk/479497/},
year = {2023},
date = {2023-07-01},
journal = {European Journal of Surgical Oncology},
abstract = {ensuremath<pensuremath>Background: Rising workflow pressures within the oesophageal cancer (OC) multidisciplinary team (MDT) can lead to variability in decision-making, and health inequality. Machine learning (ML) offers a potential automated data-driven approach to address inconsistency and standardize care. The aim of this experimental pilot study was to develop ML models able to predict curative OC MDT treatment decisions and determine the relative importance of underlying decision-critical variables. Methods: Retrospective complete-case analysis of oesophagectomy patients $pm$ neoadjuvant chemotherapy (NACT) or chemoradiotherapy (NACRT) between 2010 and 2020. Established ML algorithms (Multinomial Logistic regression (MLR), Random Forests (RF), Extreme Gradient Boosting (XGB)) and Decision Tree (DT) were used to train models predicting OC MDT treatment decisions: surgery (S), NACT + S or NACRT + S. Performance metrics included Area Under the Curve (AUC), Accuracy, Kappa, LogLoss, F1 and Precision -Recall AUC. Variable importance was calculated for each model. Results: We identified 399 cases with a male-to-female ratio of 3.6:1 and median age of 66.1yrs (range 32?83). MLR outperformed RF, XGB and DT across performance metrics (mean AUC of 0.793 [$pm$0.045] vs 0.757 [$pm$0.068], 0.740 [$pm$0.042], and 0.709 [$pm$0.021] respectively). Variable importance analysis identified age as a major factor in the decision to offer surgery alone or NACT + S across models (p < 0.05). Conclusions: ML techniques can use limited feature-sets to predict curative UGI MDT treatment decisions. Explainable Artificial Intelligence methods provide insight into decision-critical variables, highlighting underlying subconscious biases in cancer care decision-making. Such models may allow prioritization of caseload, improve efficiency, and offer data-driven decision-assistance to MDTs in the future.ensuremath</pensuremath>},
note = {Publisher Copyright:
copyright 2023 The Author(s)},
keywords = {Artificial Intelligence, machine learning, Oesophageal cancer multidisciplinary team},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Background: Rising workflow pressures within the oesophageal cancer (OC) multidisciplinary team (MDT) can lead to variability in decision-making, and health inequality. Machine learning (ML) offers a potential automated data-driven approach to address inconsistency and standardize care. The aim of this experimental pilot study was to develop ML models able to predict curative OC MDT treatment decisions and determine the relative importance of underlying decision-critical variables. Methods: Retrospective complete-case analysis of oesophagectomy patients $pm$ neoadjuvant chemotherapy (NACT) or chemoradiotherapy (NACRT) between 2010 and 2020. Established ML algorithms (Multinomial Logistic regression (MLR), Random Forests (RF), Extreme Gradient Boosting (XGB)) and Decision Tree (DT) were used to train models predicting OC MDT treatment decisions: surgery (S), NACT + S or NACRT + S. Performance metrics included Area Under the Curve (AUC), Accuracy, Kappa, LogLoss, F1 and Precision -Recall AUC. Variable importance was calculated for each model. Results: We identified 399 cases with a male-to-female ratio of 3.6:1 and median age of 66.1yrs (range 32?83). MLR outperformed RF, XGB and DT across performance metrics (mean AUC of 0.793 [$pm$0.045] vs 0.757 [$pm$0.068], 0.740 [$pm$0.042], and 0.709 [$pm$0.021] respectively). Variable importance analysis identified age as a major factor in the decision to offer surgery alone or NACT + S across models (p < 0.05). Conclusions: ML techniques can use limited feature-sets to predict curative UGI MDT treatment decisions. Explainable Artificial Intelligence methods provide insight into decision-critical variables, highlighting underlying subconscious biases in cancer care decision-making. Such models may allow prioritization of caseload, improve efficiency, and offer data-driven decision-assistance to MDTs in the future.ensuremath</pensuremath>
Abioye, Ayodeji
University of Southampton, 2023.
Abstract | Links | BibTeX | Tags:
@phdthesis{soton479472,
title = {Multimodal speech and visual gesture control interface technique for small unmanned multirotor aircraft},
author = {Ayodeji Abioye},
url = {https://eprints.soton.ac.uk/479472/},
year = {2023},
date = {2023-07-01},
publisher = {University of Southampton},
school = {University of Southampton},
abstract = {ensuremath<p class="MsoNormal"ensuremath>This research conducted an investigation into the use of novel human computer interaction(HCI) interfaces in the control of small multirotor unmanned aerial vehicles(UAVs). The main objective was to propose, design, and develop an alternative control interface for the small multirotor UAV, which could perform better than the standard RC joystick (RCJ) controller, and to evaluate the performance of the proposed interface. The multimodal speech and visual gesture (mSVG)interface were proposed, designed, and developed. This was then coupled to a Rotor S ROS Gazebo UAV simulator. An experiment study was designed to determine how practical the use of the proposed multimodal speech and visual gesture interface was in the control of small multirotor UAVs by determining the limits of speech and gesture at different ambient noise levels and under different background-lighting conditions, respectively. And to determine how the mSVG interface compares to the RC joystick controller for a simple navigational control task - in terms of performance (time of completion and accuracy of navigational control) and from a human factor?s perspective (user satisfaction and cognitive workload). 37 participants were recruited. From the results of the experiments conducted, the mSVG interface was found to be an effective alternative to the RCJ interface when operated within a constrained application environment. From the result of the noise level experiment, it was observed that speech recognition accuracy/success rate falls as noise levels rise, with75 dB noise level being the practical aerial robot (aerobot) application limit. From the results of the gesture lighting experiment, gestures were successfully recognised from 10 Lux and above on distinct solid backgrounds, but the effect of varying both the lighting conditions and the environment background on the quality of gesture recognition, was insignificant (< 0.5%), implying that the technology used, type of gesture captured, and the image processing technique used were more important. From the result of the performance and cognitive workload comparison between the RCJ and mSVG interfaces, the mSVG interface was found to perform better at higher nCA application levels than the RCJ interface. The mSVG interface was 1 minute faster and 25% more accurate than the RCJ interface; and the RCJ interface was found to be 1.4 times more cognitively demanding than the mSVG interface. The main limitation of this research was the limited lighting level range of 10 Lux - 1400 Lux used during the gesture lighting experiment, which constrains the application limit to lowlighting indoor environments. Suggested further works from this research included the development of a more robust gesture and speech algorithm and the coupling of the improved mSVG interface on to a practical UAV.ensuremath</pensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {phdthesis}
}
ensuremath<p class="MsoNormal"ensuremath>This research conducted an investigation into the use of novel human computer interaction(HCI) interfaces in the control of small multirotor unmanned aerial vehicles(UAVs). The main objective was to propose, design, and develop an alternative control interface for the small multirotor UAV, which could perform better than the standard RC joystick (RCJ) controller, and to evaluate the performance of the proposed interface. The multimodal speech and visual gesture (mSVG)interface were proposed, designed, and developed. This was then coupled to a Rotor S ROS Gazebo UAV simulator. An experiment study was designed to determine how practical the use of the proposed multimodal speech and visual gesture interface was in the control of small multirotor UAVs by determining the limits of speech and gesture at different ambient noise levels and under different background-lighting conditions, respectively. And to determine how the mSVG interface compares to the RC joystick controller for a simple navigational control task - in terms of performance (time of completion and accuracy of navigational control) and from a human factor?s perspective (user satisfaction and cognitive workload). 37 participants were recruited. From the results of the experiments conducted, the mSVG interface was found to be an effective alternative to the RCJ interface when operated within a constrained application environment. From the result of the noise level experiment, it was observed that speech recognition accuracy/success rate falls as noise levels rise, with75 dB noise level being the practical aerial robot (aerobot) application limit. From the results of the gesture lighting experiment, gestures were successfully recognised from 10 Lux and above on distinct solid backgrounds, but the effect of varying both the lighting conditions and the environment background on the quality of gesture recognition, was insignificant (< 0.5%), implying that the technology used, type of gesture captured, and the image processing technique used were more important. From the result of the performance and cognitive workload comparison between the RCJ and mSVG interfaces, the mSVG interface was found to perform better at higher nCA application levels than the RCJ interface. The mSVG interface was 1 minute faster and 25% more accurate than the RCJ interface; and the RCJ interface was found to be 1.4 times more cognitively demanding than the mSVG interface. The main limitation of this research was the limited lighting level range of 10 Lux - 1400 Lux used during the gesture lighting experiment, which constrains the application limit to lowlighting indoor environments. Suggested further works from this research included the development of a more robust gesture and speech algorithm and the coupling of the improved mSVG interface on to a practical UAV.ensuremath</pensuremath>
Abioye, Ayodeji; Naiseh, Mohammad; Hunt, William; Clark, Jediah R; Ramchurn, Sarvapali D; Soorati, Mohammad
The effect of data visualisation quality and task density on human-swarm interaction Proceedings Article
In: Proceedings of the 2023 32nd IEEE International Conference on Robot and Human Interactive Communication (RO-MAN), IEEE, 2023.
Abstract | Links | BibTeX | Tags: featured_publication
@inproceedings{soton479970,
title = {The effect of data visualisation quality and task density on human-swarm interaction},
author = {Ayodeji Abioye and Mohammad Naiseh and William Hunt and Jediah R Clark and Sarvapali D Ramchurn and Mohammad Soorati},
url = {https://eprints.soton.ac.uk/479970/},
year = {2023},
date = {2023-06-01},
urldate = {2023-06-01},
booktitle = {Proceedings of the 2023 32nd IEEE International Conference on Robot and Human Interactive Communication (RO-MAN)},
publisher = {IEEE},
abstract = {Despite the advantages of having robot swarms, human supervision is required for real-world applications. The performance of the human-swarm system depends on several factors including the data availability for the human operators. In this paper, we study the human factors aspect of the human-swarm interaction and investigate how having access to high-quality data can affect the performance of the human-swarm system - the number of tasks completed and the human trust level in operation. We designed an experiment where a human operator is tasked to operate a swarm to identify casualties in an area within a given time period. One group of operators had the option to request high-quality pictures while the other group had to base their decision on the available low-quality images. We performed a user study with 120 participants and recorded their success rate (directly logged via the simulation platform) as well as their workload and trust level (measured through a questionnaire after completing a human-swarm scenario). The findings from our study indicated that the group granted access to high-quality data exhibited an increased workload and placed greater trust in the swarm, thus confirming our initial hypothesis. However, we also found that the number of accurately identified casualties did not significantly vary between the two groups, suggesting that data quality had no impact on the successful completion of tasks.},
keywords = {featured_publication},
pubstate = {published},
tppubtype = {inproceedings}
}
Despite the advantages of having robot swarms, human supervision is required for real-world applications. The performance of the human-swarm system depends on several factors including the data availability for the human operators. In this paper, we study the human factors aspect of the human-swarm interaction and investigate how having access to high-quality data can affect the performance of the human-swarm system - the number of tasks completed and the human trust level in operation. We designed an experiment where a human operator is tasked to operate a swarm to identify casualties in an area within a given time period. One group of operators had the option to request high-quality pictures while the other group had to base their decision on the available low-quality images. We performed a user study with 120 participants and recorded their success rate (directly logged via the simulation platform) as well as their workload and trust level (measured through a questionnaire after completing a human-swarm scenario). The findings from our study indicated that the group granted access to high-quality data exhibited an increased workload and placed greater trust in the swarm, thus confirming our initial hypothesis. However, we also found that the number of accurately identified casualties did not significantly vary between the two groups, suggesting that data quality had no impact on the successful completion of tasks.
Krook, Joshua; McAuley, Derek; Anderson, Stuart; Downer, John; Winter, Peter; Ramchurn, Sarvapali D
AI Foundation Models: initial review, CMA Consultation, TAS Hub Response Miscellaneous
2023.
Links | BibTeX | Tags: Artificial Intelligence, Competition policy, featured_publication, Foundation Models, Large Language Models, markets
@misc{soton477553,
title = {AI Foundation Models: initial review, CMA Consultation, TAS Hub Response},
author = {Joshua Krook and Derek McAuley and Stuart Anderson and John Downer and Peter Winter and Sarvapali D Ramchurn},
url = {https://eprints.soton.ac.uk/477553/},
year = {2023},
date = {2023-06-01},
urldate = {2023-06-01},
publisher = {University of Southampton},
keywords = {Artificial Intelligence, Competition policy, featured_publication, Foundation Models, Large Language Models, markets},
pubstate = {published},
tppubtype = {misc}
}
Krook, Joshua; Downer, John; Winter, Peter; Williams, Jennifer; Ives, Jonathan; Bratu, Roxana; Sheir, Stephanie; Williams, Robin; Anderson, Stuart; Li, Phoebe; Ramamoorthy, Subramanian; Ramchurn, Sarvapali
AI regulation: a pro-innovation approach ? policy proposals: TASHub Response Miscellaneous
2023.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Consultation, innovation, Regulation, Trustworthy Autonomous Systems
@misc{soton478329,
title = {AI regulation: a pro-innovation approach ? policy proposals: TASHub Response},
author = {Joshua Krook and John Downer and Peter Winter and Jennifer Williams and Jonathan Ives and Roxana Bratu and Stephanie Sheir and Robin Williams and Stuart Anderson and Phoebe Li and Subramanian Ramamoorthy and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/478329/},
year = {2023},
date = {2023-06-01},
publisher = {University of Southampton},
abstract = {Response to open consultation from: Department for Science, Innovation and Technologyensuremath<br/ensuremath>and Office for Artificial Intelligence},
keywords = {Artificial Intelligence, Consultation, innovation, Regulation, Trustworthy Autonomous Systems},
pubstate = {published},
tppubtype = {misc}
}
Response to open consultation from: Department for Science, Innovation and Technologyensuremath<br/ensuremath>and Office for Artificial Intelligence
Hunt, William; Ryan, Jack; Abioye, Ayodeji O; Ramchurn, Sarvapali D; Soorati, Mohammad D
Demonstrating performance benefits of human-swarm teaming Proceedings Article
In: Proceedings of the 2023 International Conference on Autonomous Agents and Multiagent Systems, pp. 3062–3064, International Foundation for Autonomous Agents and Multiagent Systems (IFAAMAS), 2023.
Abstract | Links | BibTeX | Tags: Applications
@inproceedings{soton479903,
title = {Demonstrating performance benefits of human-swarm teaming},
author = {William Hunt and Jack Ryan and Ayodeji O Abioye and Sarvapali D Ramchurn and Mohammad D Soorati},
url = {https://eprints.soton.ac.uk/479903/},
year = {2023},
date = {2023-05-01},
urldate = {2023-05-01},
booktitle = {Proceedings of the 2023 International Conference on Autonomous Agents and Multiagent Systems},
pages = {3062–3064},
publisher = {International Foundation for Autonomous Agents and Multiagent Systems (IFAAMAS)},
abstract = {Autonomous swarms of robots can bring robustness, scalability and adaptability to safety-critical tasks such as search and rescue but their application is still very limited. Using semi-autonomous swarms with human control can bring robot swarms to real-world applications. Human operators can define goals for the swarm, monitor their performance and interfere with, or overrule, the decisions and behaviour. We present the "Human And Robot Interactive Swarm'' simulator (HARIS) that allows multi-user interaction with a robot swarm and facilitates qualitative and quantitative user studies through simulation of robot swarms completing tasks, from package delivery to search and rescue, with varying levels of human control. In this demonstration, we showcase the simulator by using it to study the performance gain offered by maintaining a "human-in-the-loop'' over a fully autonomous system as an example. This is illustrated in the context of search and rescue, with an autonomous allocation of resources to those in need.},
keywords = {Applications},
pubstate = {published},
tppubtype = {inproceedings}
}
Autonomous swarms of robots can bring robustness, scalability and adaptability to safety-critical tasks such as search and rescue but their application is still very limited. Using semi-autonomous swarms with human control can bring robot swarms to real-world applications. Human operators can define goals for the swarm, monitor their performance and interfere with, or overrule, the decisions and behaviour. We present the "Human And Robot Interactive Swarm'' simulator (HARIS) that allows multi-user interaction with a robot swarm and facilitates qualitative and quantitative user studies through simulation of robot swarms completing tasks, from package delivery to search and rescue, with varying levels of human control. In this demonstration, we showcase the simulator by using it to study the performance gain offered by maintaining a "human-in-the-loop'' over a fully autonomous system as an example. This is illustrated in the context of search and rescue, with an autonomous allocation of resources to those in need.
Worrawichaipat, Phuriwat; Gerding, Enrico; Kaparias, Ioannis; Ramchurn, Sarvapali
Multi-agent signal-less intersection management with dynamic platoon formation Proceedings Article
In: 22nd International Conference on Autonomous Agents and Multiagent Systems (29/05/23 - 02/06/23), pp. 1542–1550, 2023.
Links | BibTeX | Tags: featured_publication
@inproceedings{soton478647,
title = {Multi-agent signal-less intersection management with dynamic platoon formation},
author = {Phuriwat Worrawichaipat and Enrico Gerding and Ioannis Kaparias and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/478647/},
year = {2023},
date = {2023-05-01},
urldate = {2023-05-01},
booktitle = {22nd International Conference on Autonomous Agents and Multiagent Systems (29/05/23 - 02/06/23)},
pages = {1542–1550},
keywords = {featured_publication},
pubstate = {published},
tppubtype = {inproceedings}
}
Everett, Gregory; Beal, Ryan J; Matthews, Tim; Early, Joseph; Norman, Timothy J; Ramchurn, Sarvapali D
Inferring player location in sports matches: multi-agent spatial imputation from limited observations Miscellaneous
2023.
Abstract | Links | BibTeX | Tags: cs.LG, cs.MA
@misc{soton477020,
title = {Inferring player location in sports matches: multi-agent spatial imputation from limited observations},
author = {Gregory Everett and Ryan J Beal and Tim Matthews and Joseph Early and Timothy J Norman and Sarvapali D Ramchurn},
url = {https://eprints.soton.ac.uk/477020/},
year = {2023},
date = {2023-02-01},
abstract = {Understanding agent behaviour in Multi-Agent Systems (MAS) is an important problem in domains such as autonomous driving, disaster response, and sports analytics. Existing MAS problems typically use uniform timesteps with observations for all agents. In this work, we analyse the problem of agent location imputation, specifically posed in environments with non-uniform timesteps and limited agent observability (textttchar12695% missing values). Our approach uses Long Short-Term Memory and Graph Neural Network components to learn temporal and inter-agent patterns to predict the location of all agents at every timestep. We apply this to the domain of football (soccer) by imputing the location of all players in a game from sparse event data (e.g., shots and passes). Our model estimates player locations to within textttchar1266.9m; a textttchar12662% reduction in error from the best performing baseline. This approach facilitates downstream analysis tasks such as player physical metrics, player coverage, and team pitch control. Existing solutions to these tasks often require optical tracking data, which is expensive to obtain and only available to elite clubs. By imputing player locations from easy to obtain event data, we increase the accessibility of downstream tasks.},
keywords = {cs.LG, cs.MA},
pubstate = {published},
tppubtype = {misc}
}
Understanding agent behaviour in Multi-Agent Systems (MAS) is an important problem in domains such as autonomous driving, disaster response, and sports analytics. Existing MAS problems typically use uniform timesteps with observations for all agents. In this work, we analyse the problem of agent location imputation, specifically posed in environments with non-uniform timesteps and limited agent observability (textttchar12695% missing values). Our approach uses Long Short-Term Memory and Graph Neural Network components to learn temporal and inter-agent patterns to predict the location of all agents at every timestep. We apply this to the domain of football (soccer) by imputing the location of all players in a game from sparse event data (e.g., shots and passes). Our model estimates player locations to within textttchar1266.9m; a textttchar12662% reduction in error from the best performing baseline. This approach facilitates downstream analysis tasks such as player physical metrics, player coverage, and team pitch control. Existing solutions to these tasks often require optical tracking data, which is expensive to obtain and only available to elite clubs. By imputing player locations from easy to obtain event data, we increase the accessibility of downstream tasks.
2022
Ahmed, Sarah; Azim, Tayyaba; Early, Joseph Arthur; Ramchurn, Sarvapali
Revisiting deep fisher vectors: using fisher information to improve object classification Proceedings Article
In: British Machine Vision Conference (21/11/22 - 24/11/22), 2022.
Abstract | Links | BibTeX | Tags:
@inproceedings{soton471260,
title = {Revisiting deep fisher vectors: using fisher information to improve object classification},
author = {Sarah Ahmed and Tayyaba Azim and Joseph Arthur Early and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/471260/},
year = {2022},
date = {2022-11-01},
booktitle = {British Machine Vision Conference (21/11/22 - 24/11/22)},
abstract = {Although deep learning models have become the gold standard in achieving outstanding results on a large variety of computer vision and machine learning tasks, the use of kernel methods has still not gone out of trend because of its potential to beat deep learning performances at a number of occasions. Given the potential of kernel techniques, prior works have also proposed the use of hybrid approaches combining deep learning with kernel learning to complement their respective strengths and weaknesses. This work develops this idea further by introducing an improved version of Fisher kernels derived from the deep Boltzmann machines (DBM). Our improved deep Fisher kernel (IDFK) utilises an approximation of the Fisher information matrix to derive improved Fisher vectors. We show IDFK can be utilised to retain a high degree of class separability, making it appropriate for classification and retrieval tasks. The efficacy of the proposed approach is evaluated on three benchmark data sets: MNIST, USPS and Alphanumeric, showing an improvement in classification performance over existing kernel approaches, and comparable performance to deep learning methods, but with much reduced computational costs. Using explainable AI methods, we also demonstrate why our IDFK leads to better classification performance.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Although deep learning models have become the gold standard in achieving outstanding results on a large variety of computer vision and machine learning tasks, the use of kernel methods has still not gone out of trend because of its potential to beat deep learning performances at a number of occasions. Given the potential of kernel techniques, prior works have also proposed the use of hybrid approaches combining deep learning with kernel learning to complement their respective strengths and weaknesses. This work develops this idea further by introducing an improved version of Fisher kernels derived from the deep Boltzmann machines (DBM). Our improved deep Fisher kernel (IDFK) utilises an approximation of the Fisher information matrix to derive improved Fisher vectors. We show IDFK can be utilised to retain a high degree of class separability, making it appropriate for classification and retrieval tasks. The efficacy of the proposed approach is evaluated on three benchmark data sets: MNIST, USPS and Alphanumeric, showing an improvement in classification performance over existing kernel approaches, and comparable performance to deep learning methods, but with much reduced computational costs. Using explainable AI methods, we also demonstrate why our IDFK leads to better classification performance.
Yazdanpanah, Vahid; Gerding, Enrico; Stein, Sebastian; Dastani, Mehdi; Jonker, Catholijn M; Norman, Timothy; Ramchurn, Sarvapali
Reasoning About Responsibility in Autonomous Systems: Challenges and Opportunities Journal Article
In: AI & Society, 2022.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Citizen-Centric AI Systems, human-agent collectives, Human-Centred AI, Multiagent Responsibility Reasoning, Multiagent Systems, Trustworthy Autonomous Systems
@article{soton471971,
title = {Reasoning About Responsibility in Autonomous Systems: Challenges and Opportunities},
author = {Vahid Yazdanpanah and Enrico Gerding and Sebastian Stein and Mehdi Dastani and Catholijn M Jonker and Timothy Norman and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/471971/},
year = {2022},
date = {2022-11-01},
journal = {AI & Society},
abstract = {Ensuring the trustworthiness of autonomous systems and artificial intelligenceensuremath<br/ensuremath>is an important interdisciplinary endeavour. In this position paper, we argue thatensuremath<br/ensuremath>this endeavour will benefit from technical advancements in capturing various forms of responsibility, and we present a comprehensive research agenda to achieve this. In particular, we argue that ensuring the reliability of autonomous system can take advantage of technical approaches for quantifying degrees of responsibility and for coordinating tasks based on that. Moreover, we deem that, in certifying the legality of an AI system, formal and computationally implementable notions of responsibility, blame, accountability, and liability are applicable for addressing potential responsibility gaps (i.e., situations in which a group is responsible, but individuals? responsibility may be unclear). This is a call to enable AI systems themselves, as well as those involved in the design, monitoring, and governance of AI systems, to represent and reason about who can be seen as responsible in prospect (e.g., for completing a task in future) and who can be seen as responsible retrospectively (e.g., for a failure that has already occurred). To that end, in this work, we show that across all stages of the design, development, and deployment of Trustworthy Autonomous Systems (TAS), responsibility reasoning should play a key role. This position paper is the first step towards establishing a road-map and research agenda on how the notion of responsibility can provide novel solution concepts for ensuring the reliability and legality of TAS and, as a result, enables an effective embedding of AI technologies into society.},
keywords = {Artificial Intelligence, Citizen-Centric AI Systems, human-agent collectives, Human-Centred AI, Multiagent Responsibility Reasoning, Multiagent Systems, Trustworthy Autonomous Systems},
pubstate = {published},
tppubtype = {article}
}
Ensuring the trustworthiness of autonomous systems and artificial intelligenceensuremath<br/ensuremath>is an important interdisciplinary endeavour. In this position paper, we argue thatensuremath<br/ensuremath>this endeavour will benefit from technical advancements in capturing various forms of responsibility, and we present a comprehensive research agenda to achieve this. In particular, we argue that ensuring the reliability of autonomous system can take advantage of technical approaches for quantifying degrees of responsibility and for coordinating tasks based on that. Moreover, we deem that, in certifying the legality of an AI system, formal and computationally implementable notions of responsibility, blame, accountability, and liability are applicable for addressing potential responsibility gaps (i.e., situations in which a group is responsible, but individuals? responsibility may be unclear). This is a call to enable AI systems themselves, as well as those involved in the design, monitoring, and governance of AI systems, to represent and reason about who can be seen as responsible in prospect (e.g., for completing a task in future) and who can be seen as responsible retrospectively (e.g., for a failure that has already occurred). To that end, in this work, we show that across all stages of the design, development, and deployment of Trustworthy Autonomous Systems (TAS), responsibility reasoning should play a key role. This position paper is the first step towards establishing a road-map and research agenda on how the notion of responsibility can provide novel solution concepts for ensuring the reliability and legality of TAS and, as a result, enables an effective embedding of AI technologies into society.
Early, Joseph; Deweese, Ying-Jung; Evers, Christine; Ramchurn, Sarvapali
Scene-to-Patch earth observation: multiple instance learning for land cover classification Miscellaneous
2022, (14 pages total (4 main content; 2 acknowledgments + citations; 8 appendices); 8 figures (2 main; 6 appendix); published at "Tackling Climate Change with Machine Learning: Workshop at NeurIPS 2022").
Abstract | Links | BibTeX | Tags: cs.CV, cs.LG
@misc{soton472853,
title = {Scene-to-Patch earth observation: multiple instance learning for land cover classification},
author = {Joseph Early and Ying-Jung Deweese and Christine Evers and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/472853/},
year = {2022},
date = {2022-11-01},
abstract = {Land cover classification (LCC), and monitoring how land use changes over time, is an important process in climate change mitigation and adaptation. Existing approaches that use machine learning with Earth observation data for LCC rely on fully-annotated and segmented datasets. Creating these datasets requires a large amount of effort, and a lack of suitable datasets has become an obstacle in scaling the use of LCC. In this study, we propose Scene-to-Patch models: an alternative LCC approach utilising Multiple Instance Learning (MIL) that requires only high-level scene labels. This enables much faster development of new datasets whilst still providing segmentation through patch-level predictions, ultimately increasing the accessibility of using LCC for different scenarios. On the DeepGlobe-LCC dataset, our approach outperforms non-MIL baselines on both scene- and patch-level prediction. This work provides the foundation for expanding the use of LCC in climate change mitigation methods for technology, government, and academia.},
note = {14 pages total (4 main content; 2 acknowledgments + citations; 8 appendices); 8 figures (2 main; 6 appendix); published at "Tackling Climate Change with Machine Learning: Workshop at NeurIPS 2022"},
keywords = {cs.CV, cs.LG},
pubstate = {published},
tppubtype = {misc}
}
Land cover classification (LCC), and monitoring how land use changes over time, is an important process in climate change mitigation and adaptation. Existing approaches that use machine learning with Earth observation data for LCC rely on fully-annotated and segmented datasets. Creating these datasets requires a large amount of effort, and a lack of suitable datasets has become an obstacle in scaling the use of LCC. In this study, we propose Scene-to-Patch models: an alternative LCC approach utilising Multiple Instance Learning (MIL) that requires only high-level scene labels. This enables much faster development of new datasets whilst still providing segmentation through patch-level predictions, ultimately increasing the accessibility of using LCC for different scenarios. On the DeepGlobe-LCC dataset, our approach outperforms non-MIL baselines on both scene- and patch-level prediction. This work provides the foundation for expanding the use of LCC in climate change mitigation methods for technology, government, and academia.
Parnell, Katie; Fischer, Joel E; Clark, Jediah R; Bodenmann, Adrian; Trigo, Maria Jose Galvez; Brito, Mario; Soorati, Mohammad Divband; Plant, Katherine; Ramchurn, Sarvapali
Trustworthy UAV relationships: Applying the Schema Action World taxonomy to UAVs and UAV swarm operations Journal Article
In: International Journal of Human-Computer Interaction, 2022.
Abstract | Links | BibTeX | Tags:
@article{soton468839,
title = {Trustworthy UAV relationships: Applying the Schema Action World taxonomy to UAVs and UAV swarm operations},
author = {Katie Parnell and Joel E Fischer and Jediah R Clark and Adrian Bodenmann and Maria Jose Galvez Trigo and Mario Brito and Mohammad Divband Soorati and Katherine Plant and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/468839/},
year = {2022},
date = {2022-07-01},
journal = {International Journal of Human-Computer Interaction},
abstract = {Human Factors play a significant role inthe development and integration of avionic systems to ensure that they are trusted and can be used effectively. As Unoccupied Aerial Vehicle (UAV) technology becomes increasingly important to the aviation domain this holds true. This study aims to gain an understanding of UAV operators?trust requirements when piloting UAVs by utilising a popular aviation interview methodology (Schema World Action Research Method), in combination with key questions on trust identified from the literature. Interviews were conducted with six UAVoperators, with a range of experience. This identified the importance of past experience to trust and the expectations that operators hold. Recommendations are made that target training to inform experience, in addition to the equipment, procedures and organisational standards that can aid in developing trustworthy systems. The methodology that was developed shows promise for capturing trust within human-automation interactions},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Human Factors play a significant role inthe development and integration of avionic systems to ensure that they are trusted and can be used effectively. As Unoccupied Aerial Vehicle (UAV) technology becomes increasingly important to the aviation domain this holds true. This study aims to gain an understanding of UAV operators?trust requirements when piloting UAVs by utilising a popular aviation interview methodology (Schema World Action Research Method), in combination with key questions on trust identified from the literature. Interviews were conducted with six UAVoperators, with a range of experience. This identified the importance of past experience to trust and the expectations that operators hold. Recommendations are made that target training to inform experience, in addition to the equipment, procedures and organisational standards that can aid in developing trustworthy systems. The methodology that was developed shows promise for capturing trust within human-automation interactions
1.
Singh, Lokesh; Deshmukh, Jayati; Georgara, Athina; Nguyen, Tan Viet Tuyen; Ramchurn, Gopal
CareOps: A multi agent control room for independent living with care Proceedings Article
In: 25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26), pp. 4146–4148, 2026.
@inproceedings{soton511472,
title = {CareOps: A multi agent control room for independent living with care},
author = {Lokesh Singh and Jayati Deshmukh and Athina Georgara and Tan Viet Tuyen Nguyen and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/511472/},
year = {2026},
date = {2026-05-01},
booktitle = {25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26)},
pages = {4146–4148},
abstract = {This paper introduces CareOps, a multi-agent control-room dashboard for Independent Living with Care (ILWC) that coordinates heterogeneous sensors and a Buddy robot across multiple simulated homes. Agents fuse radar, audio, bed, passive infrared (PIR), and gait data into single prioritised incidents with human-readable explanations, and support one-click dispatch of either a robot (via socket) or a human carer (via smartphone notification). The demo presents a single decision-support workflow that orchestrates sensors, software agents, and robots through a unified interface. It provides a human-in-the-loop platform for exploring how professional care staff understand multi-sensor evidence and choose between robot and human responses.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper introduces CareOps, a multi-agent control-room dashboard for Independent Living with Care (ILWC) that coordinates heterogeneous sensors and a Buddy robot across multiple simulated homes. Agents fuse radar, audio, bed, passive infrared (PIR), and gait data into single prioritised incidents with human-readable explanations, and support one-click dispatch of either a robot (via socket) or a human carer (via smartphone notification). The demo presents a single decision-support workflow that orchestrates sensors, software agents, and robots through a unified interface. It provides a human-in-the-loop platform for exploring how professional care staff understand multi-sensor evidence and choose between robot and human responses.
2.
Georgara, Athina; Deshmukh, Jayati; Ramchurn, Gopal
GreenLine: A delay-tolerant mechanism design for grid capacity allocation Proceedings Article
In: 25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26), pp. 3323–3325, 2026.
@inproceedings{soton511473,
title = {GreenLine: A delay-tolerant mechanism design for grid capacity allocation},
author = {Athina Georgara and Jayati Deshmukh and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/511473/},
year = {2026},
date = {2026-05-01},
booktitle = {25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26)},
pages = {3323–3325},
abstract = {This work presents GreenLine, a new auction-based mechanism for integrating renewable power plants (RPP) in the power grid. We address a challenging task, where RPP owners want to maximise their profits by installing new power plants in the grid; while, the Distribution Network Operator (DNO) seeks to maximise the generated power while reducing potential installation delays due to growing power demand. This paper formulates this task as a multi-agent system, studies its useful properties such as incentive compatibility, individual rationality and economical efficiency, and discusses GreenLine under different deployment variations.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This work presents GreenLine, a new auction-based mechanism for integrating renewable power plants (RPP) in the power grid. We address a challenging task, where RPP owners want to maximise their profits by installing new power plants in the grid; while, the Distribution Network Operator (DNO) seeks to maximise the generated power while reducing potential installation delays due to growing power demand. This paper formulates this task as a multi-agent system, studies its useful properties such as incentive compatibility, individual rationality and economical efficiency, and discusses GreenLine under different deployment variations.
3.
Deshmukh, Jayati; Yazdanpanah, Vahid; Stein, Sebastian; Ramchurn, Gopal
The triad of identity, trust and responsibility in multi-agent systems Proceedings Article
In: 25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26), 2026.
@inproceedings{soton509930,
title = {The triad of identity, trust and responsibility in multi-agent systems},
author = {Jayati Deshmukh and Vahid Yazdanpanah and Sebastian Stein and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/509930/},
year = {2026},
date = {2026-05-01},
booktitle = {25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26)},
abstract = {The design of autonomous AI agents that behave responsibly and foster trust in open multi-agent systems remains a fundamental challenge. Traditional game-theoretical approaches largely assume self-interested behaviour, yet real-world collaborations among humans often rely on prosocial considerations that extend beyond individual utility. To address this, for the first time in this paper, we investigate the triad of identity, responsibility, and trust as core elements shaping responsible multi-agent behaviour. We propose a novel agent model, building on the notion of Computational Transcendence, which equips agents with an elastic sense of identity, enabling them to incorporate the welfare of others into their decision-making. Our framework integrates subjective (identity-based) and objective (experience-based and reputation-based) components of trust. Using Iterated Prisoner?s Dilemma (IPD) simulations on different network structures, we analyse how varying levels of identity and trust affect responsible behaviour. Results demonstrate that the interplay of these three concepts can promote emergent responsibility, mitigate exploitation, and sustain long-term cooperation in dynamic multi-agent environments. We argue that this triadic perspective provides a principled foundation for designing trustworthy, responsible, and identity/value aware agents with implications for future human?AI collaboration.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The design of autonomous AI agents that behave responsibly and foster trust in open multi-agent systems remains a fundamental challenge. Traditional game-theoretical approaches largely assume self-interested behaviour, yet real-world collaborations among humans often rely on prosocial considerations that extend beyond individual utility. To address this, for the first time in this paper, we investigate the triad of identity, responsibility, and trust as core elements shaping responsible multi-agent behaviour. We propose a novel agent model, building on the notion of Computational Transcendence, which equips agents with an elastic sense of identity, enabling them to incorporate the welfare of others into their decision-making. Our framework integrates subjective (identity-based) and objective (experience-based and reputation-based) components of trust. Using Iterated Prisoner?s Dilemma (IPD) simulations on different network structures, we analyse how varying levels of identity and trust affect responsible behaviour. Results demonstrate that the interplay of these three concepts can promote emergent responsibility, mitigate exploitation, and sustain long-term cooperation in dynamic multi-agent environments. We argue that this triadic perspective provides a principled foundation for designing trustworthy, responsible, and identity/value aware agents with implications for future human?AI collaboration.
4.
Everett, Gregory Alan
Models and algorithms to optimise team performance in football PhD Thesis
University of Southampton, 2026.
@phdthesis{soton509865,
title = {Models and algorithms to optimise team performance in football},
author = {Gregory Alan Everett},
url = {https://eprints.soton.ac.uk/509865/},
year = {2026},
date = {2026-03-01},
publisher = {University of Southampton},
school = {University of Southampton},
abstract = {With the growing use of Artificial Intelligence (AI) and machine learning, there is increasing potential to model and optimise team behaviour across important domains such as disaster response, security, and team sports. These domains are inherently spatiotemporal, requiring models that capture both spatial and temporal dynamics. This thesis focuses on football, a complex, dynamic sport with rich spatiotemporal data and clear objectives, making it an ideal testbed for developing and validating team-based AI models. Moreover, football analytics is a rapidly expanding industry, with European clubs generating ?38 billion in revenue during the 2023/24 season, driving the demand for models that provide competitive advantages.ensuremath<br/ensuremath>ensuremath<br/ensuremath>This thesis proposes a number of novel methods that utilise spatiotemporal data to advance team prediction, analysis, and decision-making in football and other team-based domains. In particular, we introduce a spatiotemporal agent behaviour imputation model that reduces predictive error by 62% compared to baselines in limited observability settings, significantly improving the accessibility of off-ball football analytics through imputed tracking data. We also present a novel spatial teamwork model, combining Monte Carlo tree search (MCTS) and linear programming, that optimises agent decision-making in real-world football defence, reducing opponent threat by 24%. In addition, we develop new metrics, derived from a graph attention network (GAT), to assign credit to indirect agent contributions in team-based defence. The GAT model predicts football passes with a 6% reduction in loss compared to baselines, and we show how these new metrics can greatly improve off-ball football player evaluation. Finally, we propose a dynamic team formation and agent replacement model that accounts for agent fatigue and unavailability and optimises decision-making using a multi-step MCTS algorithm. Applied to football team selection and substitutions, this model improves long-term team performance by 1% and reduces first-team injuries by 15%. This thesis also highlights key remaining challenges for AI in football, including the use of richer player data (e.g., body pose) and greater use of explainable models to build trust in clubs.},
keywords = {},
pubstate = {published},
tppubtype = {phdthesis}
}
With the growing use of Artificial Intelligence (AI) and machine learning, there is increasing potential to model and optimise team behaviour across important domains such as disaster response, security, and team sports. These domains are inherently spatiotemporal, requiring models that capture both spatial and temporal dynamics. This thesis focuses on football, a complex, dynamic sport with rich spatiotemporal data and clear objectives, making it an ideal testbed for developing and validating team-based AI models. Moreover, football analytics is a rapidly expanding industry, with European clubs generating ?38 billion in revenue during the 2023/24 season, driving the demand for models that provide competitive advantages.ensuremath<br/ensuremath>ensuremath<br/ensuremath>This thesis proposes a number of novel methods that utilise spatiotemporal data to advance team prediction, analysis, and decision-making in football and other team-based domains. In particular, we introduce a spatiotemporal agent behaviour imputation model that reduces predictive error by 62% compared to baselines in limited observability settings, significantly improving the accessibility of off-ball football analytics through imputed tracking data. We also present a novel spatial teamwork model, combining Monte Carlo tree search (MCTS) and linear programming, that optimises agent decision-making in real-world football defence, reducing opponent threat by 24%. In addition, we develop new metrics, derived from a graph attention network (GAT), to assign credit to indirect agent contributions in team-based defence. The GAT model predicts football passes with a 6% reduction in loss compared to baselines, and we show how these new metrics can greatly improve off-ball football player evaluation. Finally, we propose a dynamic team formation and agent replacement model that accounts for agent fatigue and unavailability and optimises decision-making using a multi-step MCTS algorithm. Applied to football team selection and substitutions, this model improves long-term team performance by 1% and reduces first-team injuries by 15%. This thesis also highlights key remaining challenges for AI in football, including the use of richer player data (e.g., body pose) and greater use of explainable models to build trust in clubs.
5.
Tuyen, Nguyen Tan Viet; Georgara, Athina; Singh, Lokesh; Deshmukh, Jayati; Davey, Sean; Tisdale, Paul N.; Ramchurn, Sarvapali
Supporting ?Stop and Watch? in elderly care with socially assistive robots: insights from a participatory design workshop Proceedings Article
In: Baillie, Lynne; Smart, William D.; Graaf, Maartje De; Gombolay, Matthew; Torre, Ilaria (Ed.): Companion Proceedings of the 21st ACM/IEEE International Conference on Human-Robot Interaction, HRI Companion 2026, pp. 302–306, ACM Press, 2026.
@inproceedings{soton511593,
title = {Supporting ?Stop and Watch? in elderly care with socially assistive robots: insights from a participatory design workshop},
author = {Nguyen Tan Viet Tuyen and Athina Georgara and Lokesh Singh and Jayati Deshmukh and Sean Davey and Paul N. Tisdale and Sarvapali Ramchurn},
editor = {Lynne Baillie and William D. Smart and Maartje De Graaf and Matthew Gombolay and Ilaria Torre},
url = {https://eprints.soton.ac.uk/511593/},
year = {2026},
date = {2026-03-01},
urldate = {2026-03-01},
booktitle = {Companion Proceedings of the 21st ACM/IEEE International Conference on Human-Robot Interaction, HRI Companion 2026},
pages = {302–306},
publisher = {ACM Press},
abstract = {ensuremath<pensuremath>?Stop and Watch? is an early-warning tool adopted by the UK NHS and is widely used in elderly care settings. The tool helps caregivers to recognise abnormal changes in residents? health. Despite its clinical value, the process remains highly manual, workload-intensive, and vulnerable to missed observations, particularly in environments facing staff shortages, frequent staff rotations, and increasing care demands. We argue that AI-based systems such as Socially Assistive Robots (SARs) and Assistive Living Technologies (ALTs) offer promising avenues for supporting and enhancing the ?Stop and Watch? tool. However, designing such systems requires a multidisciplinary effort to establish a comprehensive understanding of current practices and the priorities and concerns of all relevant stakeholders. This paper presents insights from a participatory design workshop held in a care home in the UK to explore how SARs and ALTs could meaningfully support the ?Stop and Watch? tool, understand stakeholders? expectations, perceived benefits, and concerns regarding deployment in this sensitive context.ensuremath</pensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
ensuremath<pensuremath>?Stop and Watch? is an early-warning tool adopted by the UK NHS and is widely used in elderly care settings. The tool helps caregivers to recognise abnormal changes in residents? health. Despite its clinical value, the process remains highly manual, workload-intensive, and vulnerable to missed observations, particularly in environments facing staff shortages, frequent staff rotations, and increasing care demands. We argue that AI-based systems such as Socially Assistive Robots (SARs) and Assistive Living Technologies (ALTs) offer promising avenues for supporting and enhancing the ?Stop and Watch? tool. However, designing such systems requires a multidisciplinary effort to establish a comprehensive understanding of current practices and the priorities and concerns of all relevant stakeholders. This paper presents insights from a participatory design workshop held in a care home in the UK to explore how SARs and ALTs could meaningfully support the ?Stop and Watch? tool, understand stakeholders? expectations, perceived benefits, and concerns regarding deployment in this sensitive context.ensuremath</pensuremath>
6.
Hafez, Muhammad Burhan; Miller, Emily; Milford, Michael J.; Ramchurn, Gopal; Ehsan, Shoaib
Through the lens of doubt: robust and efficient uncertainty estimation for visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 11, no. 5, pp. 5899–5906, 2026.
@article{soton510726,
title = {Through the lens of doubt: robust and efficient uncertainty estimation for visual place recognition},
author = {Muhammad Burhan Hafez and Emily Miller and Michael J. Milford and Gopal Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/510726/},
year = {2026},
date = {2026-03-01},
journal = {IEEE Robotics and Automation Letters},
volume = {11},
number = {5},
pages = {5899–5906},
abstract = {ensuremath<pensuremath>Visual Place Recognition (VPR) enables robots and autonomous vehicles to identify previously visited locations by matching current observations against a database of known places. However, VPR systems face significant challenges when deployed across varying visual environments, lighting conditions, seasonal changes, and viewpoints changes. Failure-critical VPR applications, such as loop closure detection in simultaneous localization and mapping (SLAM) pipelines, require robust estimation of place matching uncertainty. We propose three training-free uncertainty metrics that estimate prediction confidence by analyzing inherent statistical patterns in similarity scores from any existing VPR method. Similarity Distribution (SD) quantifies match distinctiveness by measuring score separation between candidates; Ratio Spread (RS) evaluates competitive ambiguity among top-scoring locations; and Statistical Uncertainty (SU) is a combination of SD and RS that provides a unified metric that generalizes across datasets and VPR methods without requiring validation data to select the optimal metric. All three metrics operate without additional model training, architectural modifications, or computationally expensive geometric verification. Comprehensive evaluation across nine state-of-the-art VPR methods and six benchmark datasets confirms that our metrics excel at discriminating between correct and incorrect VPR matches, and consistently outperform existing approaches while maintaining negligible computational overhead, making it deployable for real-time robotic applications across varied environmental conditions with improved precision-recall performance.ensuremath</pensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Visual Place Recognition (VPR) enables robots and autonomous vehicles to identify previously visited locations by matching current observations against a database of known places. However, VPR systems face significant challenges when deployed across varying visual environments, lighting conditions, seasonal changes, and viewpoints changes. Failure-critical VPR applications, such as loop closure detection in simultaneous localization and mapping (SLAM) pipelines, require robust estimation of place matching uncertainty. We propose three training-free uncertainty metrics that estimate prediction confidence by analyzing inherent statistical patterns in similarity scores from any existing VPR method. Similarity Distribution (SD) quantifies match distinctiveness by measuring score separation between candidates; Ratio Spread (RS) evaluates competitive ambiguity among top-scoring locations; and Statistical Uncertainty (SU) is a combination of SD and RS that provides a unified metric that generalizes across datasets and VPR methods without requiring validation data to select the optimal metric. All three metrics operate without additional model training, architectural modifications, or computationally expensive geometric verification. Comprehensive evaluation across nine state-of-the-art VPR methods and six benchmark datasets confirms that our metrics excel at discriminating between correct and incorrect VPR matches, and consistently outperform existing approaches while maintaining negligible computational overhead, making it deployable for real-time robotic applications across varied environmental conditions with improved precision-recall performance.ensuremath</pensuremath>
7.
Ramchurn, Gopal; Neff, Gina; Parisio, Isabela; Kiden, Sarah; Georgara, Athina; Stumpf, Simone; Wong, Mark; Shahandashti, Siamak F.; Aristodemou, Marios
Responsible AI UK response to the DSIT?s call for evidence on the AI Growth Lab Technical Report
no. 10.5258/RAi/005, 2026.
@techreport{soton508020,
title = {Responsible AI UK response to the DSIT?s call for evidence on the AI Growth Lab},
author = {Gopal Ramchurn and Gina Neff and Isabela Parisio and Sarah Kiden and Athina Georgara and Simone Stumpf and Mark Wong and Siamak F. Shahandashti and Marios Aristodemou},
url = {https://eprints.soton.ac.uk/508020/},
year = {2026},
date = {2026-01-01},
number = {10.5258/RAi/005},
publisher = {University of Southampton},
abstract = {A response to the Department for Science, Innovation and Technology (DSIT) open call for evidence regarding the AI Growth Lab1 on behalf of Responsible AI UK (RAi UK), an open and multidisciplinary network that brings together experts from across the four nations of the UK to understand how we should shape the development of AI to benefit people, communities and society},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
A response to the Department for Science, Innovation and Technology (DSIT) open call for evidence regarding the AI Growth Lab1 on behalf of Responsible AI UK (RAi UK), an open and multidisciplinary network that brings together experts from across the four nations of the UK to understand how we should shape the development of AI to benefit people, communities and society
8.
Owen, Aled Lloyd; Ramchurn, Gopal; Dasgupta, Prokar; Ao, Shuang; Barnard, Pepita; Deshmukh, Jayati; Parisio, Isabela; Singh, Lokesh; Valoor, Adarsh; Waheed, Maria; Braunschweig, Bertrand; Quintero, Karla; Poretschkin, Maximilian; Hawes, Ben; Khomh, Foutse
Advancing Trustworthy Artificial Intelligence: Lessons Learned and Emerging Challenges Proceedings Article
In: Responsible Ai, University of Southampton, 2025.
@inproceedings{soton506660,
title = {Advancing Trustworthy Artificial Intelligence: Lessons Learned and Emerging Challenges},
author = {Aled Lloyd Owen and Gopal Ramchurn and Prokar Dasgupta and Shuang Ao and Pepita Barnard and Jayati Deshmukh and Isabela Parisio and Lokesh Singh and Adarsh Valoor and Maria Waheed and Bertrand Braunschweig and Karla Quintero and Maximilian Poretschkin and Ben Hawes and Foutse Khomh},
url = {https://eprints.soton.ac.uk/506660/},
year = {2025},
date = {2025-11-01},
booktitle = {Responsible Ai},
publisher = {University of Southampton},
abstract = {Government white paper},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Government white paper
9.
Reyes-Cruz, Gisela; Kiden, Sarah; Azim, Tayyaba; Bergin, Aislinn Gomez; Choi, Sena; Eke, Damian; Klyshbekova, Maira; Peter, Oriane; Waheed, Maria; Devlin, Kate; Fischer, Joel; Vallejos, Elvira Perez; Ramchurn, Gopal; Stein, Sebastian
Response to call for input for EMRTD study ?Artificial Intelligence, Cultural Rights, and the Right to Development? Technical Report
no. 10.25878/7nmj-0t79, 2025.
@techreport{soton511244,
title = {Response to call for input for EMRTD study ?Artificial Intelligence, Cultural Rights, and the Right to Development?},
author = {Gisela Reyes-Cruz and Sarah Kiden and Tayyaba Azim and Aislinn Gomez Bergin and Sena Choi and Damian Eke and Maira Klyshbekova and Oriane Peter and Maria Waheed and Kate Devlin and Joel Fischer and Elvira Perez Vallejos and Gopal Ramchurn and Sebastian Stein},
url = {https://eprints.soton.ac.uk/511244/},
year = {2025},
date = {2025-11-01},
number = {10.25878/7nmj-0t79},
publisher = {University of Nottingham},
abstract = {Responsible AI UK's response to the Call for Input for EMRTD study "Artificial Intelligence, Cultural Rights, and the Right to Development" issued by the Expert Mechanism on the Right to Development United Nations Human Rights Office of the High Commissioner (OHCHR).ensuremath<br/ensuremath>ensuremath<br/ensuremath>https://www.ohchr.org/en/calls-for-input/2025/call-input-emrtd-study-artificial-intelligence-cultural-rights-and-right},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
Responsible AI UK's response to the Call for Input for EMRTD study "Artificial Intelligence, Cultural Rights, and the Right to Development" issued by the Expert Mechanism on the Right to Development United Nations Human Rights Office of the High Commissioner (OHCHR).ensuremath<br/ensuremath>ensuremath<br/ensuremath>https://www.ohchr.org/en/calls-for-input/2025/call-input-emrtd-study-artificial-intelligence-cultural-rights-and-right
10.
Ramchurn, Gopal; Owen, Aled Lloyd; Ao, Shuang; Barnard, Pepita; Bergin, Aislinn Gomez; Parisio, Isabela; Valoor, Adarsh; Waheed, Maria; Holter, Carolyn Ten; Portillo, Virginia; Procter, Rob; Batterham, Paul; Downer, John; Winter, Peter; Krook, Joshua; Blockx, Jan; Braunschweig, Bertrand; Quintero, Karla; Poretschkin, Maximilian; Becker, Daniel; Campbell-Ratcliffe, Emily; Hawes, Ben; Shaw, Patricia; Patel, Reema; Tewari, Ashish; Thomas, Alec; Duncan, Paul; Gillings, Eliot
Frameworks and Toolkits for Assuring Responsible AI Book Section
In: Responsible Ai UK, University of Southampton, 2025.
@incollection{soton506057,
title = {Frameworks and Toolkits for Assuring Responsible AI},
author = {Gopal Ramchurn and Aled Lloyd Owen and Shuang Ao and Pepita Barnard and Aislinn Gomez Bergin and Isabela Parisio and Adarsh Valoor and Maria Waheed and Carolyn Ten Holter and Virginia Portillo and Rob Procter and Paul Batterham and John Downer and Peter Winter and Joshua Krook and Jan Blockx and Bertrand Braunschweig and Karla Quintero and Maximilian Poretschkin and Daniel Becker and Emily Campbell-Ratcliffe and Ben Hawes and Patricia Shaw and Reema Patel and Ashish Tewari and Alec Thomas and Paul Duncan and Eliot Gillings},
url = {https://eprints.soton.ac.uk/506057/},
year = {2025},
date = {2025-10-01},
urldate = {2025-10-01},
booktitle = {Responsible Ai UK},
publisher = {University of Southampton},
abstract = {This report explores experiences of using tools for assuring responsible practice in the build, deployment, and governance of AI systems, in particular frameworks and toolkits developed and published by RAi UK-funded projects and by other organisations. We examine evidence of how and why these tools are used, and of their usefulness and their limitations. We look for lessons that could improve AI assurance in the future, including potentially using AI-based tools to support AI assurance. The aim is to identify priority areas and key questions for RAi UK and other researchers and organisations to explore further, with the goal of improving the fit between the supply of, and demand for, tools that support organisational assurance of responsible AI .Responsible Ai UK and our partner for this event, Confiance.ai, brought people together from across government, public services, and industry, as well as project teams that develop support for responsible AI. The aim was to clarify how people should approach the development and use of toolkits and/or frameworks in practice. We wanted to synthesise findings from our research across the programme and to find and address any gaps. The workshop took place online on Thursday 10 April 2025.},
keywords = {},
pubstate = {published},
tppubtype = {incollection}
}
This report explores experiences of using tools for assuring responsible practice in the build, deployment, and governance of AI systems, in particular frameworks and toolkits developed and published by RAi UK-funded projects and by other organisations. We examine evidence of how and why these tools are used, and of their usefulness and their limitations. We look for lessons that could improve AI assurance in the future, including potentially using AI-based tools to support AI assurance. The aim is to identify priority areas and key questions for RAi UK and other researchers and organisations to explore further, with the goal of improving the fit between the supply of, and demand for, tools that support organisational assurance of responsible AI .Responsible Ai UK and our partner for this event, Confiance.ai, brought people together from across government, public services, and industry, as well as project teams that develop support for responsible AI. The aim was to clarify how people should approach the development and use of toolkits and/or frameworks in practice. We wanted to synthesise findings from our research across the programme and to find and address any gaps. The workshop took place online on Thursday 10 April 2025.
11.
Kiden, Sarah; Peter, Oriane; Reyes-Cruz, Gisela; Klyshbekova, Maira; Choi, Sena; Bergin, Aislinn Gomez; Waheed, Maria; Eke, Damian; Azim, Tayyaba; Ramchurn, Sarvapali; Stein, Sebastian; Vallejos, Elvira Perez; Devlin, Kate; Fischer, Joel E.
Back to the communities: a mixed-methods and community-driven evaluation of cultural sensitivity in text-to-image models Miscellaneous
2025.
@misc{soton507303,
title = {Back to the communities: a mixed-methods and community-driven evaluation of cultural sensitivity in text-to-image models},
author = {Sarah Kiden and Oriane Peter and Gisela Reyes-Cruz and Maira Klyshbekova and Sena Choi and Aislinn Gomez Bergin and Maria Waheed and Damian Eke and Tayyaba Azim and Sarvapali Ramchurn and Sebastian Stein and Elvira Perez Vallejos and Kate Devlin and Joel E. Fischer},
url = {https://eprints.soton.ac.uk/507303/},
year = {2025},
date = {2025-10-01},
publisher = {arXiv},
abstract = {Evidence shows that text-to-image (T2I) models disproportionately reflect Western cultural norms, amplifying misrepresentation and harms to minority groups. However, evaluating cultural sensitivity is inherently complex due to its fluid and multifaceted nature. This paper draws on a state-of-the-art review and co-creation workshops involving 59 individuals from 19 different countries. We developed and validated a mixed-methods community-based evaluation methodology to assess cultural sensitivity in T2I models, which embraces first-person methods. Quantitative scores and qualitative inquiries expose convergence and disagreement within and across communities, illuminate the downstream consequences of misrepresentation, and trace how training data shaped by unequal power relations distort depictions. Extensive assessments are constrained by high resource requirements and the dynamic nature of culture, a tension we alleviate through a context-based and iterative methodology. The paper provides actionable recommendations for stakeholders, highlighting pathways to investigate the sources, mechanisms, and impacts of cultural (mis)representation in T2I models.},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Evidence shows that text-to-image (T2I) models disproportionately reflect Western cultural norms, amplifying misrepresentation and harms to minority groups. However, evaluating cultural sensitivity is inherently complex due to its fluid and multifaceted nature. This paper draws on a state-of-the-art review and co-creation workshops involving 59 individuals from 19 different countries. We developed and validated a mixed-methods community-based evaluation methodology to assess cultural sensitivity in T2I models, which embraces first-person methods. Quantitative scores and qualitative inquiries expose convergence and disagreement within and across communities, illuminate the downstream consequences of misrepresentation, and trace how training data shaped by unequal power relations distort depictions. Extensive assessments are constrained by high resource requirements and the dynamic nature of culture, a tension we alleviate through a context-based and iterative methodology. The paper provides actionable recommendations for stakeholders, highlighting pathways to investigate the sources, mechanisms, and impacts of cultural (mis)representation in T2I models.
12.
Ansari, Aamir Ahmad; Ramchurn, Gopal; Nguyen, Tan Viet Tuyen
Beyond text: multi-modal LLM in human robot interaction Proceedings Article
In: UK AI Research Symposium: A Festival of Ideas (08/09/25 - 09/09/25), 2025.
@inproceedings{soton505659,
title = {Beyond text: multi-modal LLM in human robot interaction},
author = {Aamir Ahmad Ansari and Gopal Ramchurn and Tan Viet Tuyen Nguyen},
url = {https://eprints.soton.ac.uk/505659/},
year = {2025},
date = {2025-09-01},
booktitle = {UK AI Research Symposium: A Festival of Ideas (08/09/25 - 09/09/25)},
abstract = {Multimodal interaction plays a vital role in Human-Robot Interaction (HRI), enabling robots to communicate with humans through multiple channels. This study introduces a novel approach to enhance such interactions by treating images and human motion as distinct foreign languages, in addition to text. In the proposed framework, vector quantization is employed to convert multimodal inputs such as images and human motions to an aligned set of tokens. A Large Language Model (LLM) is then pre-trained with the use of Low-Rank Adaptation (LoRA) and instruction-tuned on a dialogue dataset that incorporates both image and motion context. The proposed multimodal LLM framework aims to equip robots with the ability to understand and respond to complex human queries through multimodal inputs and outputs, enabling more natural and effective interactions.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Multimodal interaction plays a vital role in Human-Robot Interaction (HRI), enabling robots to communicate with humans through multiple channels. This study introduces a novel approach to enhance such interactions by treating images and human motion as distinct foreign languages, in addition to text. In the proposed framework, vector quantization is employed to convert multimodal inputs such as images and human motions to an aligned set of tokens. A Large Language Model (LLM) is then pre-trained with the use of Low-Rank Adaptation (LoRA) and instruction-tuned on a dialogue dataset that incorporates both image and motion context. The proposed multimodal LLM framework aims to equip robots with the ability to understand and respond to complex human queries through multimodal inputs and outputs, enabling more natural and effective interactions.
13.
Thavanesan, Navamayooran; Naiseh, Mohammad; Terol, Miguel; Rahman, Saqib Andrew; Hill, Samuel Luke; Parfitt, Charlotte; Walters, Zoë S; Ramchurn, Sarvapali; Markar, Sheraz; Owen, Richard; Maynard, Nick; Azim, Tayyaba; Belkatir, Zehor; Perez, Elvira Vallejos; McCord, Mimi; Underwood, Tim; Vigneswaran, Ganesh
The oesophageal cancer multi-disciplinary tool: a co-designed, externally validated, machine learning clinical decision support system Journal Article
In: EClinicalMedicine, vol. 89, pp. 103527, 2025, (For the purposes of open access, the authors have applied a Creative Commons attribution license (CC-BY) to any Author Accepted Manuscript version arising from this submission.).
@article{soton506614,
title = {The oesophageal cancer multi-disciplinary tool: a co-designed, externally validated, machine learning clinical decision support system},
author = {Navamayooran Thavanesan and Mohammad Naiseh and Miguel Terol and Saqib Andrew Rahman and Samuel Luke Hill and Charlotte Parfitt and Zoë S Walters and Sarvapali Ramchurn and Sheraz Markar and Richard Owen and Nick Maynard and Tayyaba Azim and Zehor Belkatir and Elvira Vallejos Perez and Mimi McCord and Tim Underwood and Ganesh Vigneswaran},
url = {https://eprints.soton.ac.uk/506614/},
year = {2025},
date = {2025-09-01},
journal = {EClinicalMedicine},
volume = {89},
pages = {103527},
abstract = {ensuremath<pensuremath>BACKGROUND: The oesophageal cancer (OC) multi-disciplinary team (MDT) operates under significant pressures, handling complex decision-making. Machine learning (ML) can learn complex decision-making paradigms to improve efficiency, consistency, and cost if trained and deployed responsibly. We present an externally validated ML-based clinical decision support system (CDSS) designed to predict OC MDT treatment decisions and prognosticate palliative scenarios, co-designed using Responsible Research and Innovation (RRI) principles.ensuremath</pensuremath>ensuremath<pensuremath>METHODS: Clinicopathological data collected from 1931 patients between 4th September 2009, and 8th November 2022 were used to test and validate models trained through four ML algorithms to predict curative and palliative treatment pathways along with palliative prognosis. 953 OC cases treated at University Hospitals Southampton (UHS) were used to train ML models which were externally validated on 978 OC cases from Oxford University Hospitals (OUH). Model performance was evaluated using Area Under Curve (AUC) for treatment classifiers and calibration curves for survival models. A parallel RRI program at the University of Southampton (United Kingdom) combining clinician interviews and inter-disciplinary workshops was conducted between 16.3.23 and 23.5.24. The RRI program comprised a group of 17 domain experts comprising programmers, computer scientists, clinicians and patient representatives to allow end-users to contribute towards the co-design of the CDSS user interface.ensuremath</pensuremath>ensuremath<pensuremath>FINDINGS: Cohorts differed in baseline characteristics, with the external cohort (OUH) being younger, having better performance status, and a higher prevalence of pulmonary and vascular disease. Despite these differences, on internal validation (UHS cohort) mean AUCs for the primary treatment model were: MLR 0.905 $±$ 0.048, XGB 0.909 $±$ 0.044 and RF 0.883 $±$ 0.059 (k = 5 cross-validation) and MLR 0.866 (95% CI 0.866-0.867), XGB 0.863 (0.862-0.864), RF 0.863 (0.867-0.868) on bootstrapped resampling. For the palliative classifier, mean AUCs were: MLR 0.805 $±$ 0.096, XGB 0.815 $±$ 0.081 and RF 0.793 $±$ 0.083 (k = 5 cross-validation) and MLR 0.736 (95% CI 0.734-0.737), XGB 0.799 (0.798-0.800), RF 0.781 (0.778-0.782) on bootstrapped resampling. On external validation (OUH cohort), AUCs were MLR 0.894, XGB 0.887 and RF 0.891 for the primary treatment model and MLR 0.711, XGB 0.742 and RF 0.730 for the palliative treatment classifier. Predicted survival probability from the palliative survival model was well calibrated over the first 12 months post-diagnosis in both cohorts. The RRI program provided a collaborative environment leading to valuable modifications to the CDSS including prediction explanations, visual aids for survival and integrated education for users producing a user-friendly and quick to use tool.ensuremath</pensuremath>ensuremath<pensuremath>INTERPRETATION: We present a novel, responsibly developed, externally validated AI CDSS trained to predict oesophageal cancer MDT decisions. It represents the foundations of a transformative application of ML, personalised, consistent and efficient MDT decision-support within OC which aligns to RRI principles.ensuremath</pensuremath>ensuremath<pensuremath>FUNDING: Doctoral Studentship for NT (Institute for Life Sciences (University of Southampton) & University Hospital Southampton), UKRI TAS Pump-Priming Grant (TAS_PP_00167).ensuremath</pensuremath>},
note = {For the purposes of open access, the authors have applied a Creative Commons attribution license (CC-BY) to any Author Accepted Manuscript version arising from this submission.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>BACKGROUND: The oesophageal cancer (OC) multi-disciplinary team (MDT) operates under significant pressures, handling complex decision-making. Machine learning (ML) can learn complex decision-making paradigms to improve efficiency, consistency, and cost if trained and deployed responsibly. We present an externally validated ML-based clinical decision support system (CDSS) designed to predict OC MDT treatment decisions and prognosticate palliative scenarios, co-designed using Responsible Research and Innovation (RRI) principles.ensuremath</pensuremath>ensuremath<pensuremath>METHODS: Clinicopathological data collected from 1931 patients between 4th September 2009, and 8th November 2022 were used to test and validate models trained through four ML algorithms to predict curative and palliative treatment pathways along with palliative prognosis. 953 OC cases treated at University Hospitals Southampton (UHS) were used to train ML models which were externally validated on 978 OC cases from Oxford University Hospitals (OUH). Model performance was evaluated using Area Under Curve (AUC) for treatment classifiers and calibration curves for survival models. A parallel RRI program at the University of Southampton (United Kingdom) combining clinician interviews and inter-disciplinary workshops was conducted between 16.3.23 and 23.5.24. The RRI program comprised a group of 17 domain experts comprising programmers, computer scientists, clinicians and patient representatives to allow end-users to contribute towards the co-design of the CDSS user interface.ensuremath</pensuremath>ensuremath<pensuremath>FINDINGS: Cohorts differed in baseline characteristics, with the external cohort (OUH) being younger, having better performance status, and a higher prevalence of pulmonary and vascular disease. Despite these differences, on internal validation (UHS cohort) mean AUCs for the primary treatment model were: MLR 0.905 $±$ 0.048, XGB 0.909 $±$ 0.044 and RF 0.883 $±$ 0.059 (k = 5 cross-validation) and MLR 0.866 (95% CI 0.866-0.867), XGB 0.863 (0.862-0.864), RF 0.863 (0.867-0.868) on bootstrapped resampling. For the palliative classifier, mean AUCs were: MLR 0.805 $±$ 0.096, XGB 0.815 $±$ 0.081 and RF 0.793 $±$ 0.083 (k = 5 cross-validation) and MLR 0.736 (95% CI 0.734-0.737), XGB 0.799 (0.798-0.800), RF 0.781 (0.778-0.782) on bootstrapped resampling. On external validation (OUH cohort), AUCs were MLR 0.894, XGB 0.887 and RF 0.891 for the primary treatment model and MLR 0.711, XGB 0.742 and RF 0.730 for the palliative treatment classifier. Predicted survival probability from the palliative survival model was well calibrated over the first 12 months post-diagnosis in both cohorts. The RRI program provided a collaborative environment leading to valuable modifications to the CDSS including prediction explanations, visual aids for survival and integrated education for users producing a user-friendly and quick to use tool.ensuremath</pensuremath>ensuremath<pensuremath>INTERPRETATION: We present a novel, responsibly developed, externally validated AI CDSS trained to predict oesophageal cancer MDT decisions. It represents the foundations of a transformative application of ML, personalised, consistent and efficient MDT decision-support within OC which aligns to RRI principles.ensuremath</pensuremath>ensuremath<pensuremath>FUNDING: Doctoral Studentship for NT (Institute for Life Sciences (University of Southampton) & University Hospital Southampton), UKRI TAS Pump-Priming Grant (TAS_PP_00167).ensuremath</pensuremath>
14.
Le, Thien Doanh; Nguyen, Tan Viet Tuyen; Duy, Tan Le; Ramchurn, Gopal
A multimodal large language model framework for gesture generation in social robots Proceedings Article
In: BEAR Workshop, The 2025 IEEE International Conference on Robot and Human Interactive Communication (25/08/25 - 29/08/25), 2025.
@inproceedings{soton506590,
title = {A multimodal large language model framework for gesture generation in social robots},
author = {Thien Doanh Le and Tan Viet Tuyen Nguyen and Tan Le Duy and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/506590/},
year = {2025},
date = {2025-08-01},
booktitle = {BEAR Workshop, The 2025 IEEE International Conference on Robot and Human Interactive Communication (25/08/25 - 29/08/25)},
abstract = {Non-verbal gestures play a crucial role in social robots, enabling them to signal their intentions to users during human?robot interaction (HRI). While recent research in this domain has primarily focused on robot gesture generation, there remains a limited number of studies on multimodal generation frameworks, where generated gestures are harmonized with other generated modalities, to better convey the robot?s intention to users via a wider range of communication channels. Inspired by recent advancements in multimodal large language models (MLLMs), we propose a novel framework that integrates motion generation models with existing MLLMs to produce high-quality 3D motions without the need for extensive multimodal training. Our framework comprises three key components: a Denoising Diffusion Motion Generation (DDMG) module that maps text descriptions to motion sequences using a diffusion-based approach; a Motion Decoding Alignment (MDA) module that refines motion representations by incorporating signal embeddings generated by an LLM; and a Fusion Module (FM) that integrates motion features trained from previous phases to enhance coherence and realism. We conducted a series of experiments on a publicly available dataset to evaluate the efficiency of the proposed framework in terms of motion quality, diversity, and semantic alignment. The results suggest that our multimodal approach can serve as a powerful controller for robot gesture generation, offering a more scalable and effective solution, particularly for social HRI.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Non-verbal gestures play a crucial role in social robots, enabling them to signal their intentions to users during human?robot interaction (HRI). While recent research in this domain has primarily focused on robot gesture generation, there remains a limited number of studies on multimodal generation frameworks, where generated gestures are harmonized with other generated modalities, to better convey the robot?s intention to users via a wider range of communication channels. Inspired by recent advancements in multimodal large language models (MLLMs), we propose a novel framework that integrates motion generation models with existing MLLMs to produce high-quality 3D motions without the need for extensive multimodal training. Our framework comprises three key components: a Denoising Diffusion Motion Generation (DDMG) module that maps text descriptions to motion sequences using a diffusion-based approach; a Motion Decoding Alignment (MDA) module that refines motion representations by incorporating signal embeddings generated by an LLM; and a Fusion Module (FM) that integrates motion features trained from previous phases to enhance coherence and realism. We conducted a series of experiments on a publicly available dataset to evaluate the efficiency of the proposed framework in terms of motion quality, diversity, and semantic alignment. The results suggest that our multimodal approach can serve as a powerful controller for robot gesture generation, offering a more scalable and effective solution, particularly for social HRI.
15.
Ramchurn, Gopal; Jones, Matt; Owen, Aled Lloyd; Ehsan, Shoaib; Kiden, Sarah; Eke, Damian; McStay, Andrew; Reitmaier, Thomas; Raju, Dani Kalarikalayil; Castenada, Nestor; Sailaja, Neelima; Faith, Becky; Hermann, Antony; Kalra, Kanika; Hawes, Ben; Adams, Rachel; Meier, Patrick; Sharma, Gaurav; Vishwarupe, Varad
Responsible AI. To enable flourishing in and by Low- and Middle- Income Countries (LMICs) Book Section
In: Responsible Ai UK, University of Southampton, 2025.
@incollection{soton506045,
title = {Responsible AI. To enable flourishing in and by Low- and Middle- Income Countries (LMICs)},
author = {Gopal Ramchurn and Matt Jones and Aled Lloyd Owen and Shoaib Ehsan and Sarah Kiden and Damian Eke and Andrew McStay and Thomas Reitmaier and Dani Kalarikalayil Raju and Nestor Castenada and Neelima Sailaja and Becky Faith and Antony Hermann and Kanika Kalra and Ben Hawes and Rachel Adams and Patrick Meier and Gaurav Sharma and Varad Vishwarupe},
url = {https://eprints.soton.ac.uk/506045/},
year = {2025},
date = {2025-07-01},
booktitle = {Responsible Ai UK},
publisher = {University of Southampton},
abstract = {This report explores how governments and industry in Low- and Middle-Income Countries (LMICs) might adopt AI to serve the values, needs and aspirations of individuals and communities, and drive the achievement of development goals. It draws on RAi UK projects working with partners in India and Indonesia and in countries in Africa and in South America, and identifies challenges, early solutions and policy implications. The aim is to identify priority areas and questions for RAi UK and other researchers and organisations to explore further. In this report we focus on the emerging challenges and other outputs of research on AI for LMICs. We convened leaders and members of projects that explore embedding AI tools and resources that are sensitive to local, cultural norms and end-user experience through co-creation with communities.},
keywords = {},
pubstate = {published},
tppubtype = {incollection}
}
This report explores how governments and industry in Low- and Middle-Income Countries (LMICs) might adopt AI to serve the values, needs and aspirations of individuals and communities, and drive the achievement of development goals. It draws on RAi UK projects working with partners in India and Indonesia and in countries in Africa and in South America, and identifies challenges, early solutions and policy implications. The aim is to identify priority areas and questions for RAi UK and other researchers and organisations to explore further. In this report we focus on the emerging challenges and other outputs of research on AI for LMICs. We convened leaders and members of projects that explore embedding AI tools and resources that are sensitive to local, cultural norms and end-user experience through co-creation with communities.
16.
Deshmukh, Jayati; Liang, Zijie; Yazdanpanah, Vahid; Stein, Sebastian; Ramchurn, Savapali D.
Serious games for ethical preference elicitation Proceedings Article
In: AAMAS '25: Proceedings of the 24th International Conference on Autonomous Agents and Multiagent Systems, pp. 3018 – 3020, Association for Computing Machinery, 2025.
@inproceedings{soton498743b,
title = {Serious games for ethical preference elicitation},
author = {Jayati Deshmukh and Zijie Liang and Vahid Yazdanpanah and Sebastian Stein and Savapali D. Ramchurn},
url = {https://eprints.soton.ac.uk/498743/},
year = {2025},
date = {2025-06-01},
booktitle = {AAMAS '25: Proceedings of the 24th International Conference on Autonomous Agents and Multiagent Systems},
pages = {3018 – 3020},
publisher = {Association for Computing Machinery},
abstract = {Autonomous agents acting on behalf of humans must act according to their ethical preferences. However, ethical preferences are latent and abstract and thus it is challenging to elicit them. To address this, we present a serious game that helps elicit ethical preferences in a more dynamic and engaging way than traditional methods such as questionnaires or simple dilemmas.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Autonomous agents acting on behalf of humans must act according to their ethical preferences. However, ethical preferences are latent and abstract and thus it is challenging to elicit them. To address this, we present a serious game that helps elicit ethical preferences in a more dynamic and engaging way than traditional methods such as questionnaires or simple dilemmas.
17.
Abioye, Ayodeji O.; Hunt, William; Gu, Yue; Schneiders, Eike; Naiseh, Mohammad; Archibald, Blair; Sevegnani, Michele; Ramchurn, Sarvapali D.; Fischer, Joel E.; Soorati, Mohammad D.
A user study evaluation of predictive formal modelling at runtime in human-swarm interaction Journal Article
In: ACM Transactions on Human-Robot Interaction, vol. 14, no. 4, 2025.
@article{soton501672,
title = {A user study evaluation of predictive formal modelling at runtime in human-swarm interaction},
author = {Ayodeji O. Abioye and William Hunt and Yue Gu and Eike Schneiders and Mohammad Naiseh and Blair Archibald and Michele Sevegnani and Sarvapali D. Ramchurn and Joel E. Fischer and Mohammad D. Soorati},
url = {https://eprints.soton.ac.uk/501672/},
year = {2025},
date = {2025-06-01},
journal = {ACM Transactions on Human-Robot Interaction},
volume = {14},
number = {4},
abstract = {Formal modelling is often used as part of the design and testing process of software development to ensure that components operate within suitable bounds even in unexpected circumstances. We conducted a user study evaluation of predictive formal modelling (PFM) at runtime in a human-swarm mission to determine the benefit of predictive formal modelling on performance and human-swarm interaction. 180 participants were recruited to perform the role of aerial swarm operators delivering parcels to target locations in a simulation environment. The PFM model was integrated into the simulation software to inform the operator of the estimated mission completion time given the current number of drones deployed. The operator could increase the number of parcels delivered in any time step by adding drones, which also increased costs, thus requiring the use of the minimum number of drones necessary to complete the task in the given time. We collected user feedback using standard survey questionnaires and measured performance using data obtained from the Human And Robot Interactive Swarm (HARIS) simulator. Our results show that PFM increased the performance of the human swarm team without significantly increasing the operators? workload or affecting the system's usability.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Formal modelling is often used as part of the design and testing process of software development to ensure that components operate within suitable bounds even in unexpected circumstances. We conducted a user study evaluation of predictive formal modelling (PFM) at runtime in a human-swarm mission to determine the benefit of predictive formal modelling on performance and human-swarm interaction. 180 participants were recruited to perform the role of aerial swarm operators delivering parcels to target locations in a simulation environment. The PFM model was integrated into the simulation software to inform the operator of the estimated mission completion time given the current number of drones deployed. The operator could increase the number of parcels delivered in any time step by adding drones, which also increased costs, thus requiring the use of the minimum number of drones necessary to complete the task in the given time. We collected user feedback using standard survey questionnaires and measured performance using data obtained from the Human And Robot Interactive Swarm (HARIS) simulator. Our results show that PFM increased the performance of the human swarm team without significantly increasing the operators? workload or affecting the system's usability.
18.
Deshmukh, Jayati; Liang, Zijie; Yazdanpanah, Vahid; Stein, Sebastian; Ramchurn, Sarvalpali D.
Serious games for ethical preference elicitation Proceedings Article
In: AAMAS - 2025 : The 24th International Conference on Autonomous Agents and Multiagent Systems (19/05/25 - 23/05/25), 2025.
@inproceedings{soton498743,
title = {Serious games for ethical preference elicitation},
author = {Jayati Deshmukh and Zijie Liang and Vahid Yazdanpanah and Sebastian Stein and Sarvalpali D. Ramchurn},
url = {https://eprints.soton.ac.uk/498743/},
year = {2025},
date = {2025-05-01},
booktitle = {AAMAS - 2025 : The 24th International Conference on Autonomous Agents and Multiagent Systems (19/05/25 - 23/05/25)},
abstract = {Autonomous agents acting on behalf of humans must act according to their ethical preferences. However, ethical preferences are latent and abstract and thus it is challenging to elicit them. To address this, we present a serious game that helps elicit ethical preferences in a more dynamic and engaging way than traditional methods such as questionnaires or simple dilemmas.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Autonomous agents acting on behalf of humans must act according to their ethical preferences. However, ethical preferences are latent and abstract and thus it is challenging to elicit them. To address this, we present a serious game that helps elicit ethical preferences in a more dynamic and engaging way than traditional methods such as questionnaires or simple dilemmas.
19.
Ismagilov, Timur; Ferrarini, Bruno; Milford, Michael; Tuyen, Nguyen Tan Viet; Ramchurn, Sarvapali D.; Ehsan, Shoaib
On motion blur and deblurring in visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 10, no. 5, pp. 4746–4753, 2025.
@article{soton507348,
title = {On motion blur and deblurring in visual place recognition},
author = {Timur Ismagilov and Bruno Ferrarini and Michael Milford and Nguyen Tan Viet Tuyen and Sarvapali D. Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/507348/},
year = {2025},
date = {2025-03-01},
journal = {IEEE Robotics and Automation Letters},
volume = {10},
number = {5},
pages = {4746–4753},
abstract = {ensuremath<pensuremath>Visual Place Recognition (VPR) in mobile robotics enables robots to localize themselves by recognizing previously visited locations using visual data. While the reliability of VPR methods has been extensively studied under conditions such as changes in illumination, season, weather and viewpoint, the impact of motion blur is relatively unexplored despite its relevance not only in rapid motion scenarios but also in low-light conditions where longer exposure times are necessary. Similarly, the role of image deblurring in enhancing VPR performance under motion blur has received limited attention so far. This letter bridges these gaps by introducing a new benchmark designed to evaluate VPR performance under the influence of motion blur and image deblurring. The benchmark includes three datasets that encompass a wide range of motion blur intensities, providing a comprehensive platform for analysis. Experimental results with several well-established VPR and image deblurring methods provide new insights into the effects of motion blur and the potential improvements achieved through deblurring. Building on these findings, the letter proposes adaptive deblurring strategies for VPR, designed to effectively manage motion blur in dynamic, real-world scenarios.ensuremath</pensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Visual Place Recognition (VPR) in mobile robotics enables robots to localize themselves by recognizing previously visited locations using visual data. While the reliability of VPR methods has been extensively studied under conditions such as changes in illumination, season, weather and viewpoint, the impact of motion blur is relatively unexplored despite its relevance not only in rapid motion scenarios but also in low-light conditions where longer exposure times are necessary. Similarly, the role of image deblurring in enhancing VPR performance under motion blur has received limited attention so far. This letter bridges these gaps by introducing a new benchmark designed to evaluate VPR performance under the influence of motion blur and image deblurring. The benchmark includes three datasets that encompass a wide range of motion blur intensities, providing a comprehensive platform for analysis. Experimental results with several well-established VPR and image deblurring methods provide new insights into the effects of motion blur and the potential improvements achieved through deblurring. Building on these findings, the letter proposes adaptive deblurring strategies for VPR, designed to effectively manage motion blur in dynamic, real-world scenarios.ensuremath</pensuremath>
20.
Abioye, Ayodeji; Hunt, William; Schneiders, Eike; Gu, Yue; Naiseh, Mohammad; Archibald, Blair; Sevegnani, Michele; Ramchurn, Gopal; Fischer, Joel; Soorati, Mohammad
Research Data for Paper: A User Study Evaluation of Predictive Formal Modelling at Runtime in Human-Swarm Interaction Miscellaneous
2025.
@misc{soton500788,
title = {Research Data for Paper: A User Study Evaluation of Predictive Formal Modelling at Runtime in Human-Swarm Interaction},
author = {Ayodeji Abioye and William Hunt and Eike Schneiders and Yue Gu and Mohammad Naiseh and Blair Archibald and Michele Sevegnani and Gopal Ramchurn and Joel Fischer and Mohammad Soorati},
url = {https://eprints.soton.ac.uk/500788/},
year = {2025},
date = {2025-03-01},
publisher = {figshare},
abstract = {ensuremath<span style='white-space:pre-line'ensuremath>This dataset is for the user study conducted to evaluate the performance benefits of predictive formal modelling (PFM) at runtime in a human-swarm interaction experiment. We recruited 180 participants to perform the role of aerial swarm operators conducting a drone delivery mission in a simulation environment using the Human And Robot Interactive Swarm (HARIS) simulator.ensuremath</spanensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
ensuremath<span style='white-space:pre-line'ensuremath>This dataset is for the user study conducted to evaluate the performance benefits of predictive formal modelling (PFM) at runtime in a human-swarm interaction experiment. We recruited 180 participants to perform the role of aerial swarm operators conducting a drone delivery mission in a simulation environment using the Human And Robot Interactive Swarm (HARIS) simulator.ensuremath</spanensuremath>
2026
Deshmukh, Jayati; Yazdanpanah, Vahid; Stein, Sebastian; Ramchurn, Gopal
The triad of identity, trust and responsibility in multi-agent systems Proceedings Article
In: 25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26), 2026.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Coordination, Ethics, Identity, Multiagent Systems, Norms, Responsibility, Trust
@inproceedings{soton509930,
title = {The triad of identity, trust and responsibility in multi-agent systems},
author = {Jayati Deshmukh and Vahid Yazdanpanah and Sebastian Stein and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/509930/},
year = {2026},
date = {2026-05-01},
booktitle = {25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26)},
abstract = {The design of autonomous AI agents that behave responsibly and foster trust in open multi-agent systems remains a fundamental challenge. Traditional game-theoretical approaches largely assume self-interested behaviour, yet real-world collaborations among humans often rely on prosocial considerations that extend beyond individual utility. To address this, for the first time in this paper, we investigate the triad of identity, responsibility, and trust as core elements shaping responsible multi-agent behaviour. We propose a novel agent model, building on the notion of Computational Transcendence, which equips agents with an elastic sense of identity, enabling them to incorporate the welfare of others into their decision-making. Our framework integrates subjective (identity-based) and objective (experience-based and reputation-based) components of trust. Using Iterated Prisoner?s Dilemma (IPD) simulations on different network structures, we analyse how varying levels of identity and trust affect responsible behaviour. Results demonstrate that the interplay of these three concepts can promote emergent responsibility, mitigate exploitation, and sustain long-term cooperation in dynamic multi-agent environments. We argue that this triadic perspective provides a principled foundation for designing trustworthy, responsible, and identity/value aware agents with implications for future human?AI collaboration.},
keywords = {Artificial Intelligence, Coordination, Ethics, Identity, Multiagent Systems, Norms, Responsibility, Trust},
pubstate = {published},
tppubtype = {inproceedings}
}
The design of autonomous AI agents that behave responsibly and foster trust in open multi-agent systems remains a fundamental challenge. Traditional game-theoretical approaches largely assume self-interested behaviour, yet real-world collaborations among humans often rely on prosocial considerations that extend beyond individual utility. To address this, for the first time in this paper, we investigate the triad of identity, responsibility, and trust as core elements shaping responsible multi-agent behaviour. We propose a novel agent model, building on the notion of Computational Transcendence, which equips agents with an elastic sense of identity, enabling them to incorporate the welfare of others into their decision-making. Our framework integrates subjective (identity-based) and objective (experience-based and reputation-based) components of trust. Using Iterated Prisoner?s Dilemma (IPD) simulations on different network structures, we analyse how varying levels of identity and trust affect responsible behaviour. Results demonstrate that the interplay of these three concepts can promote emergent responsibility, mitigate exploitation, and sustain long-term cooperation in dynamic multi-agent environments. We argue that this triadic perspective provides a principled foundation for designing trustworthy, responsible, and identity/value aware agents with implications for future human?AI collaboration.
Georgara, Athina; Deshmukh, Jayati; Ramchurn, Gopal
GreenLine: A delay-tolerant mechanism design for grid capacity allocation Proceedings Article
In: 25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26), pp. 3323–3325, 2026.
Abstract | Links | BibTeX | Tags: auction, mechanism design, power grid, resource allocation
@inproceedings{soton511473,
title = {GreenLine: A delay-tolerant mechanism design for grid capacity allocation},
author = {Athina Georgara and Jayati Deshmukh and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/511473/},
year = {2026},
date = {2026-05-01},
booktitle = {25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26)},
pages = {3323–3325},
abstract = {This work presents GreenLine, a new auction-based mechanism for integrating renewable power plants (RPP) in the power grid. We address a challenging task, where RPP owners want to maximise their profits by installing new power plants in the grid; while, the Distribution Network Operator (DNO) seeks to maximise the generated power while reducing potential installation delays due to growing power demand. This paper formulates this task as a multi-agent system, studies its useful properties such as incentive compatibility, individual rationality and economical efficiency, and discusses GreenLine under different deployment variations.},
keywords = {auction, mechanism design, power grid, resource allocation},
pubstate = {published},
tppubtype = {inproceedings}
}
This work presents GreenLine, a new auction-based mechanism for integrating renewable power plants (RPP) in the power grid. We address a challenging task, where RPP owners want to maximise their profits by installing new power plants in the grid; while, the Distribution Network Operator (DNO) seeks to maximise the generated power while reducing potential installation delays due to growing power demand. This paper formulates this task as a multi-agent system, studies its useful properties such as incentive compatibility, individual rationality and economical efficiency, and discusses GreenLine under different deployment variations.
Singh, Lokesh; Deshmukh, Jayati; Georgara, Athina; Nguyen, Tan Viet Tuyen; Ramchurn, Gopal
CareOps: A multi agent control room for independent living with care Proceedings Article
In: 25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26), pp. 4146–4148, 2026.
Abstract | Links | BibTeX | Tags: Assistive robotics, Human-robot interaction, Independent living, Multi-agent Coordination
@inproceedings{soton511472,
title = {CareOps: A multi agent control room for independent living with care},
author = {Lokesh Singh and Jayati Deshmukh and Athina Georgara and Tan Viet Tuyen Nguyen and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/511472/},
year = {2026},
date = {2026-05-01},
booktitle = {25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26)},
pages = {4146–4148},
abstract = {This paper introduces CareOps, a multi-agent control-room dashboard for Independent Living with Care (ILWC) that coordinates heterogeneous sensors and a Buddy robot across multiple simulated homes. Agents fuse radar, audio, bed, passive infrared (PIR), and gait data into single prioritised incidents with human-readable explanations, and support one-click dispatch of either a robot (via socket) or a human carer (via smartphone notification). The demo presents a single decision-support workflow that orchestrates sensors, software agents, and robots through a unified interface. It provides a human-in-the-loop platform for exploring how professional care staff understand multi-sensor evidence and choose between robot and human responses.},
keywords = {Assistive robotics, Human-robot interaction, Independent living, Multi-agent Coordination},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper introduces CareOps, a multi-agent control-room dashboard for Independent Living with Care (ILWC) that coordinates heterogeneous sensors and a Buddy robot across multiple simulated homes. Agents fuse radar, audio, bed, passive infrared (PIR), and gait data into single prioritised incidents with human-readable explanations, and support one-click dispatch of either a robot (via socket) or a human carer (via smartphone notification). The demo presents a single decision-support workflow that orchestrates sensors, software agents, and robots through a unified interface. It provides a human-in-the-loop platform for exploring how professional care staff understand multi-sensor evidence and choose between robot and human responses.
Hafez, Muhammad Burhan; Miller, Emily; Milford, Michael J.; Ramchurn, Gopal; Ehsan, Shoaib
Through the lens of doubt: robust and efficient uncertainty estimation for visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 11, no. 5, pp. 5899–5906, 2026.
Abstract | Links | BibTeX | Tags: Deep Learning for Visual Perception, Localization, Vision-Based Navigation
@article{soton510726,
title = {Through the lens of doubt: robust and efficient uncertainty estimation for visual place recognition},
author = {Muhammad Burhan Hafez and Emily Miller and Michael J. Milford and Gopal Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/510726/},
year = {2026},
date = {2026-03-01},
journal = {IEEE Robotics and Automation Letters},
volume = {11},
number = {5},
pages = {5899–5906},
abstract = {ensuremath<pensuremath>Visual Place Recognition (VPR) enables robots and autonomous vehicles to identify previously visited locations by matching current observations against a database of known places. However, VPR systems face significant challenges when deployed across varying visual environments, lighting conditions, seasonal changes, and viewpoints changes. Failure-critical VPR applications, such as loop closure detection in simultaneous localization and mapping (SLAM) pipelines, require robust estimation of place matching uncertainty. We propose three training-free uncertainty metrics that estimate prediction confidence by analyzing inherent statistical patterns in similarity scores from any existing VPR method. Similarity Distribution (SD) quantifies match distinctiveness by measuring score separation between candidates; Ratio Spread (RS) evaluates competitive ambiguity among top-scoring locations; and Statistical Uncertainty (SU) is a combination of SD and RS that provides a unified metric that generalizes across datasets and VPR methods without requiring validation data to select the optimal metric. All three metrics operate without additional model training, architectural modifications, or computationally expensive geometric verification. Comprehensive evaluation across nine state-of-the-art VPR methods and six benchmark datasets confirms that our metrics excel at discriminating between correct and incorrect VPR matches, and consistently outperform existing approaches while maintaining negligible computational overhead, making it deployable for real-time robotic applications across varied environmental conditions with improved precision-recall performance.ensuremath</pensuremath>},
keywords = {Deep Learning for Visual Perception, Localization, Vision-Based Navigation},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Visual Place Recognition (VPR) enables robots and autonomous vehicles to identify previously visited locations by matching current observations against a database of known places. However, VPR systems face significant challenges when deployed across varying visual environments, lighting conditions, seasonal changes, and viewpoints changes. Failure-critical VPR applications, such as loop closure detection in simultaneous localization and mapping (SLAM) pipelines, require robust estimation of place matching uncertainty. We propose three training-free uncertainty metrics that estimate prediction confidence by analyzing inherent statistical patterns in similarity scores from any existing VPR method. Similarity Distribution (SD) quantifies match distinctiveness by measuring score separation between candidates; Ratio Spread (RS) evaluates competitive ambiguity among top-scoring locations; and Statistical Uncertainty (SU) is a combination of SD and RS that provides a unified metric that generalizes across datasets and VPR methods without requiring validation data to select the optimal metric. All three metrics operate without additional model training, architectural modifications, or computationally expensive geometric verification. Comprehensive evaluation across nine state-of-the-art VPR methods and six benchmark datasets confirms that our metrics excel at discriminating between correct and incorrect VPR matches, and consistently outperform existing approaches while maintaining negligible computational overhead, making it deployable for real-time robotic applications across varied environmental conditions with improved precision-recall performance.ensuremath</pensuremath>
Tuyen, Nguyen Tan Viet; Georgara, Athina; Singh, Lokesh; Deshmukh, Jayati; Davey, Sean; Tisdale, Paul N.; Ramchurn, Sarvapali
Supporting ?Stop and Watch? in elderly care with socially assistive robots: insights from a participatory design workshop Proceedings Article
In: Baillie, Lynne; Smart, William D.; Graaf, Maartje De; Gombolay, Matthew; Torre, Ilaria (Ed.): Companion Proceedings of the 21st ACM/IEEE International Conference on Human-Robot Interaction, HRI Companion 2026, pp. 302–306, ACM Press, 2026.
Abstract | Links | BibTeX | Tags: assistive living technologies, elderly care, featured_publication, socially assistive robots
@inproceedings{soton511593,
title = {Supporting ?Stop and Watch? in elderly care with socially assistive robots: insights from a participatory design workshop},
author = {Nguyen Tan Viet Tuyen and Athina Georgara and Lokesh Singh and Jayati Deshmukh and Sean Davey and Paul N. Tisdale and Sarvapali Ramchurn},
editor = {Lynne Baillie and William D. Smart and Maartje De Graaf and Matthew Gombolay and Ilaria Torre},
url = {https://eprints.soton.ac.uk/511593/},
year = {2026},
date = {2026-03-01},
urldate = {2026-03-01},
booktitle = {Companion Proceedings of the 21st ACM/IEEE International Conference on Human-Robot Interaction, HRI Companion 2026},
pages = {302–306},
publisher = {ACM Press},
abstract = {ensuremath<pensuremath>?Stop and Watch? is an early-warning tool adopted by the UK NHS and is widely used in elderly care settings. The tool helps caregivers to recognise abnormal changes in residents? health. Despite its clinical value, the process remains highly manual, workload-intensive, and vulnerable to missed observations, particularly in environments facing staff shortages, frequent staff rotations, and increasing care demands. We argue that AI-based systems such as Socially Assistive Robots (SARs) and Assistive Living Technologies (ALTs) offer promising avenues for supporting and enhancing the ?Stop and Watch? tool. However, designing such systems requires a multidisciplinary effort to establish a comprehensive understanding of current practices and the priorities and concerns of all relevant stakeholders. This paper presents insights from a participatory design workshop held in a care home in the UK to explore how SARs and ALTs could meaningfully support the ?Stop and Watch? tool, understand stakeholders? expectations, perceived benefits, and concerns regarding deployment in this sensitive context.ensuremath</pensuremath>},
keywords = {assistive living technologies, elderly care, featured_publication, socially assistive robots},
pubstate = {published},
tppubtype = {inproceedings}
}
ensuremath<pensuremath>?Stop and Watch? is an early-warning tool adopted by the UK NHS and is widely used in elderly care settings. The tool helps caregivers to recognise abnormal changes in residents? health. Despite its clinical value, the process remains highly manual, workload-intensive, and vulnerable to missed observations, particularly in environments facing staff shortages, frequent staff rotations, and increasing care demands. We argue that AI-based systems such as Socially Assistive Robots (SARs) and Assistive Living Technologies (ALTs) offer promising avenues for supporting and enhancing the ?Stop and Watch? tool. However, designing such systems requires a multidisciplinary effort to establish a comprehensive understanding of current practices and the priorities and concerns of all relevant stakeholders. This paper presents insights from a participatory design workshop held in a care home in the UK to explore how SARs and ALTs could meaningfully support the ?Stop and Watch? tool, understand stakeholders? expectations, perceived benefits, and concerns regarding deployment in this sensitive context.ensuremath</pensuremath>
Everett, Gregory Alan
Models and algorithms to optimise team performance in football PhD Thesis
University of Southampton, 2026.
Abstract | Links | BibTeX | Tags: applied machine learning, machine learning, multi-agent systems, sports analytics, team optimisation
@phdthesis{soton509865,
title = {Models and algorithms to optimise team performance in football},
author = {Gregory Alan Everett},
url = {https://eprints.soton.ac.uk/509865/},
year = {2026},
date = {2026-03-01},
publisher = {University of Southampton},
school = {University of Southampton},
abstract = {With the growing use of Artificial Intelligence (AI) and machine learning, there is increasing potential to model and optimise team behaviour across important domains such as disaster response, security, and team sports. These domains are inherently spatiotemporal, requiring models that capture both spatial and temporal dynamics. This thesis focuses on football, a complex, dynamic sport with rich spatiotemporal data and clear objectives, making it an ideal testbed for developing and validating team-based AI models. Moreover, football analytics is a rapidly expanding industry, with European clubs generating ?38 billion in revenue during the 2023/24 season, driving the demand for models that provide competitive advantages.ensuremath<br/ensuremath>ensuremath<br/ensuremath>This thesis proposes a number of novel methods that utilise spatiotemporal data to advance team prediction, analysis, and decision-making in football and other team-based domains. In particular, we introduce a spatiotemporal agent behaviour imputation model that reduces predictive error by 62% compared to baselines in limited observability settings, significantly improving the accessibility of off-ball football analytics through imputed tracking data. We also present a novel spatial teamwork model, combining Monte Carlo tree search (MCTS) and linear programming, that optimises agent decision-making in real-world football defence, reducing opponent threat by 24%. In addition, we develop new metrics, derived from a graph attention network (GAT), to assign credit to indirect agent contributions in team-based defence. The GAT model predicts football passes with a 6% reduction in loss compared to baselines, and we show how these new metrics can greatly improve off-ball football player evaluation. Finally, we propose a dynamic team formation and agent replacement model that accounts for agent fatigue and unavailability and optimises decision-making using a multi-step MCTS algorithm. Applied to football team selection and substitutions, this model improves long-term team performance by 1% and reduces first-team injuries by 15%. This thesis also highlights key remaining challenges for AI in football, including the use of richer player data (e.g., body pose) and greater use of explainable models to build trust in clubs.},
keywords = {applied machine learning, machine learning, multi-agent systems, sports analytics, team optimisation},
pubstate = {published},
tppubtype = {phdthesis}
}
With the growing use of Artificial Intelligence (AI) and machine learning, there is increasing potential to model and optimise team behaviour across important domains such as disaster response, security, and team sports. These domains are inherently spatiotemporal, requiring models that capture both spatial and temporal dynamics. This thesis focuses on football, a complex, dynamic sport with rich spatiotemporal data and clear objectives, making it an ideal testbed for developing and validating team-based AI models. Moreover, football analytics is a rapidly expanding industry, with European clubs generating ?38 billion in revenue during the 2023/24 season, driving the demand for models that provide competitive advantages.ensuremath<br/ensuremath>ensuremath<br/ensuremath>This thesis proposes a number of novel methods that utilise spatiotemporal data to advance team prediction, analysis, and decision-making in football and other team-based domains. In particular, we introduce a spatiotemporal agent behaviour imputation model that reduces predictive error by 62% compared to baselines in limited observability settings, significantly improving the accessibility of off-ball football analytics through imputed tracking data. We also present a novel spatial teamwork model, combining Monte Carlo tree search (MCTS) and linear programming, that optimises agent decision-making in real-world football defence, reducing opponent threat by 24%. In addition, we develop new metrics, derived from a graph attention network (GAT), to assign credit to indirect agent contributions in team-based defence. The GAT model predicts football passes with a 6% reduction in loss compared to baselines, and we show how these new metrics can greatly improve off-ball football player evaluation. Finally, we propose a dynamic team formation and agent replacement model that accounts for agent fatigue and unavailability and optimises decision-making using a multi-step MCTS algorithm. Applied to football team selection and substitutions, this model improves long-term team performance by 1% and reduces first-team injuries by 15%. This thesis also highlights key remaining challenges for AI in football, including the use of richer player data (e.g., body pose) and greater use of explainable models to build trust in clubs.
Ramchurn, Gopal; Neff, Gina; Parisio, Isabela; Kiden, Sarah; Georgara, Athina; Stumpf, Simone; Wong, Mark; Shahandashti, Siamak F.; Aristodemou, Marios
Responsible AI UK response to the DSIT?s call for evidence on the AI Growth Lab Technical Report
no. 10.5258/RAi/005, 2026.
Abstract | Links | BibTeX | Tags:
@techreport{soton508020,
title = {Responsible AI UK response to the DSIT?s call for evidence on the AI Growth Lab},
author = {Gopal Ramchurn and Gina Neff and Isabela Parisio and Sarah Kiden and Athina Georgara and Simone Stumpf and Mark Wong and Siamak F. Shahandashti and Marios Aristodemou},
url = {https://eprints.soton.ac.uk/508020/},
year = {2026},
date = {2026-01-01},
number = {10.5258/RAi/005},
publisher = {University of Southampton},
abstract = {A response to the Department for Science, Innovation and Technology (DSIT) open call for evidence regarding the AI Growth Lab1 on behalf of Responsible AI UK (RAi UK), an open and multidisciplinary network that brings together experts from across the four nations of the UK to understand how we should shape the development of AI to benefit people, communities and society},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
A response to the Department for Science, Innovation and Technology (DSIT) open call for evidence regarding the AI Growth Lab1 on behalf of Responsible AI UK (RAi UK), an open and multidisciplinary network that brings together experts from across the four nations of the UK to understand how we should shape the development of AI to benefit people, communities and society
2025
Reyes-Cruz, Gisela; Kiden, Sarah; Azim, Tayyaba; Bergin, Aislinn Gomez; Choi, Sena; Eke, Damian; Klyshbekova, Maira; Peter, Oriane; Waheed, Maria; Devlin, Kate; Fischer, Joel; Vallejos, Elvira Perez; Ramchurn, Gopal; Stein, Sebastian
Response to call for input for EMRTD study ?Artificial Intelligence, Cultural Rights, and the Right to Development? Technical Report
no. 10.25878/7nmj-0t79, 2025.
Abstract | Links | BibTeX | Tags:
@techreport{soton511244,
title = {Response to call for input for EMRTD study ?Artificial Intelligence, Cultural Rights, and the Right to Development?},
author = {Gisela Reyes-Cruz and Sarah Kiden and Tayyaba Azim and Aislinn Gomez Bergin and Sena Choi and Damian Eke and Maira Klyshbekova and Oriane Peter and Maria Waheed and Kate Devlin and Joel Fischer and Elvira Perez Vallejos and Gopal Ramchurn and Sebastian Stein},
url = {https://eprints.soton.ac.uk/511244/},
year = {2025},
date = {2025-11-01},
number = {10.25878/7nmj-0t79},
publisher = {University of Nottingham},
abstract = {Responsible AI UK's response to the Call for Input for EMRTD study "Artificial Intelligence, Cultural Rights, and the Right to Development" issued by the Expert Mechanism on the Right to Development United Nations Human Rights Office of the High Commissioner (OHCHR).ensuremath<br/ensuremath>ensuremath<br/ensuremath>https://www.ohchr.org/en/calls-for-input/2025/call-input-emrtd-study-artificial-intelligence-cultural-rights-and-right},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
Responsible AI UK's response to the Call for Input for EMRTD study "Artificial Intelligence, Cultural Rights, and the Right to Development" issued by the Expert Mechanism on the Right to Development United Nations Human Rights Office of the High Commissioner (OHCHR).ensuremath<br/ensuremath>ensuremath<br/ensuremath>https://www.ohchr.org/en/calls-for-input/2025/call-input-emrtd-study-artificial-intelligence-cultural-rights-and-right
Owen, Aled Lloyd; Ramchurn, Gopal; Dasgupta, Prokar; Ao, Shuang; Barnard, Pepita; Deshmukh, Jayati; Parisio, Isabela; Singh, Lokesh; Valoor, Adarsh; Waheed, Maria; Braunschweig, Bertrand; Quintero, Karla; Poretschkin, Maximilian; Hawes, Ben; Khomh, Foutse
Advancing Trustworthy Artificial Intelligence: Lessons Learned and Emerging Challenges Proceedings Article
In: Responsible Ai, University of Southampton, 2025.
Abstract | Links | BibTeX | Tags:
@inproceedings{soton506660,
title = {Advancing Trustworthy Artificial Intelligence: Lessons Learned and Emerging Challenges},
author = {Aled Lloyd Owen and Gopal Ramchurn and Prokar Dasgupta and Shuang Ao and Pepita Barnard and Jayati Deshmukh and Isabela Parisio and Lokesh Singh and Adarsh Valoor and Maria Waheed and Bertrand Braunschweig and Karla Quintero and Maximilian Poretschkin and Ben Hawes and Foutse Khomh},
url = {https://eprints.soton.ac.uk/506660/},
year = {2025},
date = {2025-11-01},
booktitle = {Responsible Ai},
publisher = {University of Southampton},
abstract = {Government white paper},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Government white paper
Kiden, Sarah; Peter, Oriane; Reyes-Cruz, Gisela; Klyshbekova, Maira; Choi, Sena; Bergin, Aislinn Gomez; Waheed, Maria; Eke, Damian; Azim, Tayyaba; Ramchurn, Sarvapali; Stein, Sebastian; Vallejos, Elvira Perez; Devlin, Kate; Fischer, Joel E.
Back to the communities: a mixed-methods and community-driven evaluation of cultural sensitivity in text-to-image models Miscellaneous
2025.
Abstract | Links | BibTeX | Tags: cs.CY, cs.SI
@misc{soton507303,
title = {Back to the communities: a mixed-methods and community-driven evaluation of cultural sensitivity in text-to-image models},
author = {Sarah Kiden and Oriane Peter and Gisela Reyes-Cruz and Maira Klyshbekova and Sena Choi and Aislinn Gomez Bergin and Maria Waheed and Damian Eke and Tayyaba Azim and Sarvapali Ramchurn and Sebastian Stein and Elvira Perez Vallejos and Kate Devlin and Joel E. Fischer},
url = {https://eprints.soton.ac.uk/507303/},
year = {2025},
date = {2025-10-01},
publisher = {arXiv},
abstract = {Evidence shows that text-to-image (T2I) models disproportionately reflect Western cultural norms, amplifying misrepresentation and harms to minority groups. However, evaluating cultural sensitivity is inherently complex due to its fluid and multifaceted nature. This paper draws on a state-of-the-art review and co-creation workshops involving 59 individuals from 19 different countries. We developed and validated a mixed-methods community-based evaluation methodology to assess cultural sensitivity in T2I models, which embraces first-person methods. Quantitative scores and qualitative inquiries expose convergence and disagreement within and across communities, illuminate the downstream consequences of misrepresentation, and trace how training data shaped by unequal power relations distort depictions. Extensive assessments are constrained by high resource requirements and the dynamic nature of culture, a tension we alleviate through a context-based and iterative methodology. The paper provides actionable recommendations for stakeholders, highlighting pathways to investigate the sources, mechanisms, and impacts of cultural (mis)representation in T2I models.},
keywords = {cs.CY, cs.SI},
pubstate = {published},
tppubtype = {misc}
}
Evidence shows that text-to-image (T2I) models disproportionately reflect Western cultural norms, amplifying misrepresentation and harms to minority groups. However, evaluating cultural sensitivity is inherently complex due to its fluid and multifaceted nature. This paper draws on a state-of-the-art review and co-creation workshops involving 59 individuals from 19 different countries. We developed and validated a mixed-methods community-based evaluation methodology to assess cultural sensitivity in T2I models, which embraces first-person methods. Quantitative scores and qualitative inquiries expose convergence and disagreement within and across communities, illuminate the downstream consequences of misrepresentation, and trace how training data shaped by unequal power relations distort depictions. Extensive assessments are constrained by high resource requirements and the dynamic nature of culture, a tension we alleviate through a context-based and iterative methodology. The paper provides actionable recommendations for stakeholders, highlighting pathways to investigate the sources, mechanisms, and impacts of cultural (mis)representation in T2I models.
Ramchurn, Gopal; Owen, Aled Lloyd; Ao, Shuang; Barnard, Pepita; Bergin, Aislinn Gomez; Parisio, Isabela; Valoor, Adarsh; Waheed, Maria; Holter, Carolyn Ten; Portillo, Virginia; Procter, Rob; Batterham, Paul; Downer, John; Winter, Peter; Krook, Joshua; Blockx, Jan; Braunschweig, Bertrand; Quintero, Karla; Poretschkin, Maximilian; Becker, Daniel; Campbell-Ratcliffe, Emily; Hawes, Ben; Shaw, Patricia; Patel, Reema; Tewari, Ashish; Thomas, Alec; Duncan, Paul; Gillings, Eliot
Frameworks and Toolkits for Assuring Responsible AI Book Section
In: Responsible Ai UK, University of Southampton, 2025.
Abstract | Links | BibTeX | Tags: featured_publication
@incollection{soton506057,
title = {Frameworks and Toolkits for Assuring Responsible AI},
author = {Gopal Ramchurn and Aled Lloyd Owen and Shuang Ao and Pepita Barnard and Aislinn Gomez Bergin and Isabela Parisio and Adarsh Valoor and Maria Waheed and Carolyn Ten Holter and Virginia Portillo and Rob Procter and Paul Batterham and John Downer and Peter Winter and Joshua Krook and Jan Blockx and Bertrand Braunschweig and Karla Quintero and Maximilian Poretschkin and Daniel Becker and Emily Campbell-Ratcliffe and Ben Hawes and Patricia Shaw and Reema Patel and Ashish Tewari and Alec Thomas and Paul Duncan and Eliot Gillings},
url = {https://eprints.soton.ac.uk/506057/},
year = {2025},
date = {2025-10-01},
urldate = {2025-10-01},
booktitle = {Responsible Ai UK},
publisher = {University of Southampton},
abstract = {This report explores experiences of using tools for assuring responsible practice in the build, deployment, and governance of AI systems, in particular frameworks and toolkits developed and published by RAi UK-funded projects and by other organisations. We examine evidence of how and why these tools are used, and of their usefulness and their limitations. We look for lessons that could improve AI assurance in the future, including potentially using AI-based tools to support AI assurance. The aim is to identify priority areas and key questions for RAi UK and other researchers and organisations to explore further, with the goal of improving the fit between the supply of, and demand for, tools that support organisational assurance of responsible AI .Responsible Ai UK and our partner for this event, Confiance.ai, brought people together from across government, public services, and industry, as well as project teams that develop support for responsible AI. The aim was to clarify how people should approach the development and use of toolkits and/or frameworks in practice. We wanted to synthesise findings from our research across the programme and to find and address any gaps. The workshop took place online on Thursday 10 April 2025.},
keywords = {featured_publication},
pubstate = {published},
tppubtype = {incollection}
}
This report explores experiences of using tools for assuring responsible practice in the build, deployment, and governance of AI systems, in particular frameworks and toolkits developed and published by RAi UK-funded projects and by other organisations. We examine evidence of how and why these tools are used, and of their usefulness and their limitations. We look for lessons that could improve AI assurance in the future, including potentially using AI-based tools to support AI assurance. The aim is to identify priority areas and key questions for RAi UK and other researchers and organisations to explore further, with the goal of improving the fit between the supply of, and demand for, tools that support organisational assurance of responsible AI .Responsible Ai UK and our partner for this event, Confiance.ai, brought people together from across government, public services, and industry, as well as project teams that develop support for responsible AI. The aim was to clarify how people should approach the development and use of toolkits and/or frameworks in practice. We wanted to synthesise findings from our research across the programme and to find and address any gaps. The workshop took place online on Thursday 10 April 2025.
Thavanesan, Navamayooran; Naiseh, Mohammad; Terol, Miguel; Rahman, Saqib Andrew; Hill, Samuel Luke; Parfitt, Charlotte; Walters, Zoë S; Ramchurn, Sarvapali; Markar, Sheraz; Owen, Richard; Maynard, Nick; Azim, Tayyaba; Belkatir, Zehor; Perez, Elvira Vallejos; McCord, Mimi; Underwood, Tim; Vigneswaran, Ganesh
The oesophageal cancer multi-disciplinary tool: a co-designed, externally validated, machine learning clinical decision support system Journal Article
In: EClinicalMedicine, vol. 89, pp. 103527, 2025, (For the purposes of open access, the authors have applied a Creative Commons attribution license (CC-BY) to any Author Accepted Manuscript version arising from this submission.).
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Decision support tool, machine learning, MDT, Multidisciplinary teams, Oesophageal cancer
@article{soton506614,
title = {The oesophageal cancer multi-disciplinary tool: a co-designed, externally validated, machine learning clinical decision support system},
author = {Navamayooran Thavanesan and Mohammad Naiseh and Miguel Terol and Saqib Andrew Rahman and Samuel Luke Hill and Charlotte Parfitt and Zoë S Walters and Sarvapali Ramchurn and Sheraz Markar and Richard Owen and Nick Maynard and Tayyaba Azim and Zehor Belkatir and Elvira Vallejos Perez and Mimi McCord and Tim Underwood and Ganesh Vigneswaran},
url = {https://eprints.soton.ac.uk/506614/},
year = {2025},
date = {2025-09-01},
journal = {EClinicalMedicine},
volume = {89},
pages = {103527},
abstract = {ensuremath<pensuremath>BACKGROUND: The oesophageal cancer (OC) multi-disciplinary team (MDT) operates under significant pressures, handling complex decision-making. Machine learning (ML) can learn complex decision-making paradigms to improve efficiency, consistency, and cost if trained and deployed responsibly. We present an externally validated ML-based clinical decision support system (CDSS) designed to predict OC MDT treatment decisions and prognosticate palliative scenarios, co-designed using Responsible Research and Innovation (RRI) principles.ensuremath</pensuremath>ensuremath<pensuremath>METHODS: Clinicopathological data collected from 1931 patients between 4th September 2009, and 8th November 2022 were used to test and validate models trained through four ML algorithms to predict curative and palliative treatment pathways along with palliative prognosis. 953 OC cases treated at University Hospitals Southampton (UHS) were used to train ML models which were externally validated on 978 OC cases from Oxford University Hospitals (OUH). Model performance was evaluated using Area Under Curve (AUC) for treatment classifiers and calibration curves for survival models. A parallel RRI program at the University of Southampton (United Kingdom) combining clinician interviews and inter-disciplinary workshops was conducted between 16.3.23 and 23.5.24. The RRI program comprised a group of 17 domain experts comprising programmers, computer scientists, clinicians and patient representatives to allow end-users to contribute towards the co-design of the CDSS user interface.ensuremath</pensuremath>ensuremath<pensuremath>FINDINGS: Cohorts differed in baseline characteristics, with the external cohort (OUH) being younger, having better performance status, and a higher prevalence of pulmonary and vascular disease. Despite these differences, on internal validation (UHS cohort) mean AUCs for the primary treatment model were: MLR 0.905 $±$ 0.048, XGB 0.909 $±$ 0.044 and RF 0.883 $±$ 0.059 (k = 5 cross-validation) and MLR 0.866 (95% CI 0.866-0.867), XGB 0.863 (0.862-0.864), RF 0.863 (0.867-0.868) on bootstrapped resampling. For the palliative classifier, mean AUCs were: MLR 0.805 $±$ 0.096, XGB 0.815 $±$ 0.081 and RF 0.793 $±$ 0.083 (k = 5 cross-validation) and MLR 0.736 (95% CI 0.734-0.737), XGB 0.799 (0.798-0.800), RF 0.781 (0.778-0.782) on bootstrapped resampling. On external validation (OUH cohort), AUCs were MLR 0.894, XGB 0.887 and RF 0.891 for the primary treatment model and MLR 0.711, XGB 0.742 and RF 0.730 for the palliative treatment classifier. Predicted survival probability from the palliative survival model was well calibrated over the first 12 months post-diagnosis in both cohorts. The RRI program provided a collaborative environment leading to valuable modifications to the CDSS including prediction explanations, visual aids for survival and integrated education for users producing a user-friendly and quick to use tool.ensuremath</pensuremath>ensuremath<pensuremath>INTERPRETATION: We present a novel, responsibly developed, externally validated AI CDSS trained to predict oesophageal cancer MDT decisions. It represents the foundations of a transformative application of ML, personalised, consistent and efficient MDT decision-support within OC which aligns to RRI principles.ensuremath</pensuremath>ensuremath<pensuremath>FUNDING: Doctoral Studentship for NT (Institute for Life Sciences (University of Southampton) & University Hospital Southampton), UKRI TAS Pump-Priming Grant (TAS_PP_00167).ensuremath</pensuremath>},
note = {For the purposes of open access, the authors have applied a Creative Commons attribution license (CC-BY) to any Author Accepted Manuscript version arising from this submission.},
keywords = {Artificial Intelligence, Decision support tool, machine learning, MDT, Multidisciplinary teams, Oesophageal cancer},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>BACKGROUND: The oesophageal cancer (OC) multi-disciplinary team (MDT) operates under significant pressures, handling complex decision-making. Machine learning (ML) can learn complex decision-making paradigms to improve efficiency, consistency, and cost if trained and deployed responsibly. We present an externally validated ML-based clinical decision support system (CDSS) designed to predict OC MDT treatment decisions and prognosticate palliative scenarios, co-designed using Responsible Research and Innovation (RRI) principles.ensuremath</pensuremath>ensuremath<pensuremath>METHODS: Clinicopathological data collected from 1931 patients between 4th September 2009, and 8th November 2022 were used to test and validate models trained through four ML algorithms to predict curative and palliative treatment pathways along with palliative prognosis. 953 OC cases treated at University Hospitals Southampton (UHS) were used to train ML models which were externally validated on 978 OC cases from Oxford University Hospitals (OUH). Model performance was evaluated using Area Under Curve (AUC) for treatment classifiers and calibration curves for survival models. A parallel RRI program at the University of Southampton (United Kingdom) combining clinician interviews and inter-disciplinary workshops was conducted between 16.3.23 and 23.5.24. The RRI program comprised a group of 17 domain experts comprising programmers, computer scientists, clinicians and patient representatives to allow end-users to contribute towards the co-design of the CDSS user interface.ensuremath</pensuremath>ensuremath<pensuremath>FINDINGS: Cohorts differed in baseline characteristics, with the external cohort (OUH) being younger, having better performance status, and a higher prevalence of pulmonary and vascular disease. Despite these differences, on internal validation (UHS cohort) mean AUCs for the primary treatment model were: MLR 0.905 $±$ 0.048, XGB 0.909 $±$ 0.044 and RF 0.883 $±$ 0.059 (k = 5 cross-validation) and MLR 0.866 (95% CI 0.866-0.867), XGB 0.863 (0.862-0.864), RF 0.863 (0.867-0.868) on bootstrapped resampling. For the palliative classifier, mean AUCs were: MLR 0.805 $±$ 0.096, XGB 0.815 $±$ 0.081 and RF 0.793 $±$ 0.083 (k = 5 cross-validation) and MLR 0.736 (95% CI 0.734-0.737), XGB 0.799 (0.798-0.800), RF 0.781 (0.778-0.782) on bootstrapped resampling. On external validation (OUH cohort), AUCs were MLR 0.894, XGB 0.887 and RF 0.891 for the primary treatment model and MLR 0.711, XGB 0.742 and RF 0.730 for the palliative treatment classifier. Predicted survival probability from the palliative survival model was well calibrated over the first 12 months post-diagnosis in both cohorts. The RRI program provided a collaborative environment leading to valuable modifications to the CDSS including prediction explanations, visual aids for survival and integrated education for users producing a user-friendly and quick to use tool.ensuremath</pensuremath>ensuremath<pensuremath>INTERPRETATION: We present a novel, responsibly developed, externally validated AI CDSS trained to predict oesophageal cancer MDT decisions. It represents the foundations of a transformative application of ML, personalised, consistent and efficient MDT decision-support within OC which aligns to RRI principles.ensuremath</pensuremath>ensuremath<pensuremath>FUNDING: Doctoral Studentship for NT (Institute for Life Sciences (University of Southampton) & University Hospital Southampton), UKRI TAS Pump-Priming Grant (TAS_PP_00167).ensuremath</pensuremath>
Ansari, Aamir Ahmad; Ramchurn, Gopal; Nguyen, Tan Viet Tuyen
Beyond text: multi-modal LLM in human robot interaction Proceedings Article
In: UK AI Research Symposium: A Festival of Ideas (08/09/25 - 09/09/25), 2025.
Abstract | Links | BibTeX | Tags:
@inproceedings{soton505659,
title = {Beyond text: multi-modal LLM in human robot interaction},
author = {Aamir Ahmad Ansari and Gopal Ramchurn and Tan Viet Tuyen Nguyen},
url = {https://eprints.soton.ac.uk/505659/},
year = {2025},
date = {2025-09-01},
booktitle = {UK AI Research Symposium: A Festival of Ideas (08/09/25 - 09/09/25)},
abstract = {Multimodal interaction plays a vital role in Human-Robot Interaction (HRI), enabling robots to communicate with humans through multiple channels. This study introduces a novel approach to enhance such interactions by treating images and human motion as distinct foreign languages, in addition to text. In the proposed framework, vector quantization is employed to convert multimodal inputs such as images and human motions to an aligned set of tokens. A Large Language Model (LLM) is then pre-trained with the use of Low-Rank Adaptation (LoRA) and instruction-tuned on a dialogue dataset that incorporates both image and motion context. The proposed multimodal LLM framework aims to equip robots with the ability to understand and respond to complex human queries through multimodal inputs and outputs, enabling more natural and effective interactions.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Multimodal interaction plays a vital role in Human-Robot Interaction (HRI), enabling robots to communicate with humans through multiple channels. This study introduces a novel approach to enhance such interactions by treating images and human motion as distinct foreign languages, in addition to text. In the proposed framework, vector quantization is employed to convert multimodal inputs such as images and human motions to an aligned set of tokens. A Large Language Model (LLM) is then pre-trained with the use of Low-Rank Adaptation (LoRA) and instruction-tuned on a dialogue dataset that incorporates both image and motion context. The proposed multimodal LLM framework aims to equip robots with the ability to understand and respond to complex human queries through multimodal inputs and outputs, enabling more natural and effective interactions.
Le, Thien Doanh; Nguyen, Tan Viet Tuyen; Duy, Tan Le; Ramchurn, Gopal
A multimodal large language model framework for gesture generation in social robots Proceedings Article
In: BEAR Workshop, The 2025 IEEE International Conference on Robot and Human Interactive Communication (25/08/25 - 29/08/25), 2025.
Abstract | Links | BibTeX | Tags:
@inproceedings{soton506590,
title = {A multimodal large language model framework for gesture generation in social robots},
author = {Thien Doanh Le and Tan Viet Tuyen Nguyen and Tan Le Duy and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/506590/},
year = {2025},
date = {2025-08-01},
booktitle = {BEAR Workshop, The 2025 IEEE International Conference on Robot and Human Interactive Communication (25/08/25 - 29/08/25)},
abstract = {Non-verbal gestures play a crucial role in social robots, enabling them to signal their intentions to users during human?robot interaction (HRI). While recent research in this domain has primarily focused on robot gesture generation, there remains a limited number of studies on multimodal generation frameworks, where generated gestures are harmonized with other generated modalities, to better convey the robot?s intention to users via a wider range of communication channels. Inspired by recent advancements in multimodal large language models (MLLMs), we propose a novel framework that integrates motion generation models with existing MLLMs to produce high-quality 3D motions without the need for extensive multimodal training. Our framework comprises three key components: a Denoising Diffusion Motion Generation (DDMG) module that maps text descriptions to motion sequences using a diffusion-based approach; a Motion Decoding Alignment (MDA) module that refines motion representations by incorporating signal embeddings generated by an LLM; and a Fusion Module (FM) that integrates motion features trained from previous phases to enhance coherence and realism. We conducted a series of experiments on a publicly available dataset to evaluate the efficiency of the proposed framework in terms of motion quality, diversity, and semantic alignment. The results suggest that our multimodal approach can serve as a powerful controller for robot gesture generation, offering a more scalable and effective solution, particularly for social HRI.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Non-verbal gestures play a crucial role in social robots, enabling them to signal their intentions to users during human?robot interaction (HRI). While recent research in this domain has primarily focused on robot gesture generation, there remains a limited number of studies on multimodal generation frameworks, where generated gestures are harmonized with other generated modalities, to better convey the robot?s intention to users via a wider range of communication channels. Inspired by recent advancements in multimodal large language models (MLLMs), we propose a novel framework that integrates motion generation models with existing MLLMs to produce high-quality 3D motions without the need for extensive multimodal training. Our framework comprises three key components: a Denoising Diffusion Motion Generation (DDMG) module that maps text descriptions to motion sequences using a diffusion-based approach; a Motion Decoding Alignment (MDA) module that refines motion representations by incorporating signal embeddings generated by an LLM; and a Fusion Module (FM) that integrates motion features trained from previous phases to enhance coherence and realism. We conducted a series of experiments on a publicly available dataset to evaluate the efficiency of the proposed framework in terms of motion quality, diversity, and semantic alignment. The results suggest that our multimodal approach can serve as a powerful controller for robot gesture generation, offering a more scalable and effective solution, particularly for social HRI.
Ramchurn, Gopal; Jones, Matt; Owen, Aled Lloyd; Ehsan, Shoaib; Kiden, Sarah; Eke, Damian; McStay, Andrew; Reitmaier, Thomas; Raju, Dani Kalarikalayil; Castenada, Nestor; Sailaja, Neelima; Faith, Becky; Hermann, Antony; Kalra, Kanika; Hawes, Ben; Adams, Rachel; Meier, Patrick; Sharma, Gaurav; Vishwarupe, Varad
Responsible AI. To enable flourishing in and by Low- and Middle- Income Countries (LMICs) Book Section
In: Responsible Ai UK, University of Southampton, 2025.
Abstract | Links | BibTeX | Tags:
@incollection{soton506045,
title = {Responsible AI. To enable flourishing in and by Low- and Middle- Income Countries (LMICs)},
author = {Gopal Ramchurn and Matt Jones and Aled Lloyd Owen and Shoaib Ehsan and Sarah Kiden and Damian Eke and Andrew McStay and Thomas Reitmaier and Dani Kalarikalayil Raju and Nestor Castenada and Neelima Sailaja and Becky Faith and Antony Hermann and Kanika Kalra and Ben Hawes and Rachel Adams and Patrick Meier and Gaurav Sharma and Varad Vishwarupe},
url = {https://eprints.soton.ac.uk/506045/},
year = {2025},
date = {2025-07-01},
booktitle = {Responsible Ai UK},
publisher = {University of Southampton},
abstract = {This report explores how governments and industry in Low- and Middle-Income Countries (LMICs) might adopt AI to serve the values, needs and aspirations of individuals and communities, and drive the achievement of development goals. It draws on RAi UK projects working with partners in India and Indonesia and in countries in Africa and in South America, and identifies challenges, early solutions and policy implications. The aim is to identify priority areas and questions for RAi UK and other researchers and organisations to explore further. In this report we focus on the emerging challenges and other outputs of research on AI for LMICs. We convened leaders and members of projects that explore embedding AI tools and resources that are sensitive to local, cultural norms and end-user experience through co-creation with communities.},
keywords = {},
pubstate = {published},
tppubtype = {incollection}
}
This report explores how governments and industry in Low- and Middle-Income Countries (LMICs) might adopt AI to serve the values, needs and aspirations of individuals and communities, and drive the achievement of development goals. It draws on RAi UK projects working with partners in India and Indonesia and in countries in Africa and in South America, and identifies challenges, early solutions and policy implications. The aim is to identify priority areas and questions for RAi UK and other researchers and organisations to explore further. In this report we focus on the emerging challenges and other outputs of research on AI for LMICs. We convened leaders and members of projects that explore embedding AI tools and resources that are sensitive to local, cultural norms and end-user experience through co-creation with communities.
Abioye, Ayodeji O.; Hunt, William; Gu, Yue; Schneiders, Eike; Naiseh, Mohammad; Archibald, Blair; Sevegnani, Michele; Ramchurn, Sarvapali D.; Fischer, Joel E.; Soorati, Mohammad D.
A user study evaluation of predictive formal modelling at runtime in human-swarm interaction Journal Article
In: ACM Transactions on Human-Robot Interaction, vol. 14, no. 4, 2025.
Abstract | Links | BibTeX | Tags: Additional Key Words and PhrasesHuman-Robot Interaction (HRI), Human-Swarm Interaction (HSI), Predictive Formal Modelling (PFM), Usability, Workload
@article{soton501672,
title = {A user study evaluation of predictive formal modelling at runtime in human-swarm interaction},
author = {Ayodeji O. Abioye and William Hunt and Yue Gu and Eike Schneiders and Mohammad Naiseh and Blair Archibald and Michele Sevegnani and Sarvapali D. Ramchurn and Joel E. Fischer and Mohammad D. Soorati},
url = {https://eprints.soton.ac.uk/501672/},
year = {2025},
date = {2025-06-01},
journal = {ACM Transactions on Human-Robot Interaction},
volume = {14},
number = {4},
abstract = {Formal modelling is often used as part of the design and testing process of software development to ensure that components operate within suitable bounds even in unexpected circumstances. We conducted a user study evaluation of predictive formal modelling (PFM) at runtime in a human-swarm mission to determine the benefit of predictive formal modelling on performance and human-swarm interaction. 180 participants were recruited to perform the role of aerial swarm operators delivering parcels to target locations in a simulation environment. The PFM model was integrated into the simulation software to inform the operator of the estimated mission completion time given the current number of drones deployed. The operator could increase the number of parcels delivered in any time step by adding drones, which also increased costs, thus requiring the use of the minimum number of drones necessary to complete the task in the given time. We collected user feedback using standard survey questionnaires and measured performance using data obtained from the Human And Robot Interactive Swarm (HARIS) simulator. Our results show that PFM increased the performance of the human swarm team without significantly increasing the operators? workload or affecting the system's usability.},
keywords = {Additional Key Words and PhrasesHuman-Robot Interaction (HRI), Human-Swarm Interaction (HSI), Predictive Formal Modelling (PFM), Usability, Workload},
pubstate = {published},
tppubtype = {article}
}
Formal modelling is often used as part of the design and testing process of software development to ensure that components operate within suitable bounds even in unexpected circumstances. We conducted a user study evaluation of predictive formal modelling (PFM) at runtime in a human-swarm mission to determine the benefit of predictive formal modelling on performance and human-swarm interaction. 180 participants were recruited to perform the role of aerial swarm operators delivering parcels to target locations in a simulation environment. The PFM model was integrated into the simulation software to inform the operator of the estimated mission completion time given the current number of drones deployed. The operator could increase the number of parcels delivered in any time step by adding drones, which also increased costs, thus requiring the use of the minimum number of drones necessary to complete the task in the given time. We collected user feedback using standard survey questionnaires and measured performance using data obtained from the Human And Robot Interactive Swarm (HARIS) simulator. Our results show that PFM increased the performance of the human swarm team without significantly increasing the operators? workload or affecting the system's usability.
Deshmukh, Jayati; Liang, Zijie; Yazdanpanah, Vahid; Stein, Sebastian; Ramchurn, Savapali D.
Serious games for ethical preference elicitation Proceedings Article
In: AAMAS '25: Proceedings of the 24th International Conference on Autonomous Agents and Multiagent Systems, pp. 3018 – 3020, Association for Computing Machinery, 2025.
Abstract | Links | BibTeX | Tags:
@inproceedings{soton498743b,
title = {Serious games for ethical preference elicitation},
author = {Jayati Deshmukh and Zijie Liang and Vahid Yazdanpanah and Sebastian Stein and Savapali D. Ramchurn},
url = {https://eprints.soton.ac.uk/498743/},
year = {2025},
date = {2025-06-01},
booktitle = {AAMAS '25: Proceedings of the 24th International Conference on Autonomous Agents and Multiagent Systems},
pages = {3018 – 3020},
publisher = {Association for Computing Machinery},
abstract = {Autonomous agents acting on behalf of humans must act according to their ethical preferences. However, ethical preferences are latent and abstract and thus it is challenging to elicit them. To address this, we present a serious game that helps elicit ethical preferences in a more dynamic and engaging way than traditional methods such as questionnaires or simple dilemmas.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Autonomous agents acting on behalf of humans must act according to their ethical preferences. However, ethical preferences are latent and abstract and thus it is challenging to elicit them. To address this, we present a serious game that helps elicit ethical preferences in a more dynamic and engaging way than traditional methods such as questionnaires or simple dilemmas.
Deshmukh, Jayati; Liang, Zijie; Yazdanpanah, Vahid; Stein, Sebastian; Ramchurn, Sarvalpali D.
Serious games for ethical preference elicitation Proceedings Article
In: AAMAS - 2025 : The 24th International Conference on Autonomous Agents and Multiagent Systems (19/05/25 - 23/05/25), 2025.
Abstract | Links | BibTeX | Tags:
@inproceedings{soton498743,
title = {Serious games for ethical preference elicitation},
author = {Jayati Deshmukh and Zijie Liang and Vahid Yazdanpanah and Sebastian Stein and Sarvalpali D. Ramchurn},
url = {https://eprints.soton.ac.uk/498743/},
year = {2025},
date = {2025-05-01},
booktitle = {AAMAS - 2025 : The 24th International Conference on Autonomous Agents and Multiagent Systems (19/05/25 - 23/05/25)},
abstract = {Autonomous agents acting on behalf of humans must act according to their ethical preferences. However, ethical preferences are latent and abstract and thus it is challenging to elicit them. To address this, we present a serious game that helps elicit ethical preferences in a more dynamic and engaging way than traditional methods such as questionnaires or simple dilemmas.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Autonomous agents acting on behalf of humans must act according to their ethical preferences. However, ethical preferences are latent and abstract and thus it is challenging to elicit them. To address this, we present a serious game that helps elicit ethical preferences in a more dynamic and engaging way than traditional methods such as questionnaires or simple dilemmas.
Abioye, Ayodeji; Hunt, William; Schneiders, Eike; Gu, Yue; Naiseh, Mohammad; Archibald, Blair; Sevegnani, Michele; Ramchurn, Gopal; Fischer, Joel; Soorati, Mohammad
Research Data for Paper: A User Study Evaluation of Predictive Formal Modelling at Runtime in Human-Swarm Interaction Miscellaneous
2025.
Abstract | Links | BibTeX | Tags:
@misc{soton500788,
title = {Research Data for Paper: A User Study Evaluation of Predictive Formal Modelling at Runtime in Human-Swarm Interaction},
author = {Ayodeji Abioye and William Hunt and Eike Schneiders and Yue Gu and Mohammad Naiseh and Blair Archibald and Michele Sevegnani and Gopal Ramchurn and Joel Fischer and Mohammad Soorati},
url = {https://eprints.soton.ac.uk/500788/},
year = {2025},
date = {2025-03-01},
publisher = {figshare},
abstract = {ensuremath<span style='white-space:pre-line'ensuremath>This dataset is for the user study conducted to evaluate the performance benefits of predictive formal modelling (PFM) at runtime in a human-swarm interaction experiment. We recruited 180 participants to perform the role of aerial swarm operators conducting a drone delivery mission in a simulation environment using the Human And Robot Interactive Swarm (HARIS) simulator.ensuremath</spanensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
ensuremath<span style='white-space:pre-line'ensuremath>This dataset is for the user study conducted to evaluate the performance benefits of predictive formal modelling (PFM) at runtime in a human-swarm interaction experiment. We recruited 180 participants to perform the role of aerial swarm operators conducting a drone delivery mission in a simulation environment using the Human And Robot Interactive Swarm (HARIS) simulator.ensuremath</spanensuremath>
Ismagilov, Timur; Ferrarini, Bruno; Milford, Michael; Tuyen, Nguyen Tan Viet; Ramchurn, Sarvapali D.; Ehsan, Shoaib
On motion blur and deblurring in visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 10, no. 5, pp. 4746–4753, 2025.
Abstract | Links | BibTeX | Tags: data sets for robotic vision, Localization, Vision-Based Navigation
@article{soton507348,
title = {On motion blur and deblurring in visual place recognition},
author = {Timur Ismagilov and Bruno Ferrarini and Michael Milford and Nguyen Tan Viet Tuyen and Sarvapali D. Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/507348/},
year = {2025},
date = {2025-03-01},
journal = {IEEE Robotics and Automation Letters},
volume = {10},
number = {5},
pages = {4746–4753},
abstract = {ensuremath<pensuremath>Visual Place Recognition (VPR) in mobile robotics enables robots to localize themselves by recognizing previously visited locations using visual data. While the reliability of VPR methods has been extensively studied under conditions such as changes in illumination, season, weather and viewpoint, the impact of motion blur is relatively unexplored despite its relevance not only in rapid motion scenarios but also in low-light conditions where longer exposure times are necessary. Similarly, the role of image deblurring in enhancing VPR performance under motion blur has received limited attention so far. This letter bridges these gaps by introducing a new benchmark designed to evaluate VPR performance under the influence of motion blur and image deblurring. The benchmark includes three datasets that encompass a wide range of motion blur intensities, providing a comprehensive platform for analysis. Experimental results with several well-established VPR and image deblurring methods provide new insights into the effects of motion blur and the potential improvements achieved through deblurring. Building on these findings, the letter proposes adaptive deblurring strategies for VPR, designed to effectively manage motion blur in dynamic, real-world scenarios.ensuremath</pensuremath>},
keywords = {data sets for robotic vision, Localization, Vision-Based Navigation},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Visual Place Recognition (VPR) in mobile robotics enables robots to localize themselves by recognizing previously visited locations using visual data. While the reliability of VPR methods has been extensively studied under conditions such as changes in illumination, season, weather and viewpoint, the impact of motion blur is relatively unexplored despite its relevance not only in rapid motion scenarios but also in low-light conditions where longer exposure times are necessary. Similarly, the role of image deblurring in enhancing VPR performance under motion blur has received limited attention so far. This letter bridges these gaps by introducing a new benchmark designed to evaluate VPR performance under the influence of motion blur and image deblurring. The benchmark includes three datasets that encompass a wide range of motion blur intensities, providing a comprehensive platform for analysis. Experimental results with several well-established VPR and image deblurring methods provide new insights into the effects of motion blur and the potential improvements achieved through deblurring. Building on these findings, the letter proposes adaptive deblurring strategies for VPR, designed to effectively manage motion blur in dynamic, real-world scenarios.ensuremath</pensuremath>
Grainge, Oliver Edward; Milford, Michael J.; Bodala, Indu; Ramchurn, Sarvapali D.; Ehsan, Shoaib
Structured pruning for efficient visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 10, no. 2, pp. 2024–2031, 2025.
Abstract | Links | BibTeX | Tags: accuracy, computational modelling, convolutional neural network, feature extraction, Localization, memory management, Real-time systems, representation learning, robustness, Vision-Based Navigation, visual place recognition (VPR), visualization
@article{soton502843,
title = {Structured pruning for efficient visual place recognition},
author = {Oliver Edward Grainge and Michael J. Milford and Indu Bodala and Sarvapali D. Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/502843/},
year = {2025},
date = {2025-02-01},
journal = {IEEE Robotics and Automation Letters},
volume = {10},
number = {2},
pages = {2024–2031},
abstract = {Visual Place Recognition (VPR) is fundamental for the global re-localization of robots and devices, enabling them to recognize previously visited locations based on visual inputs. This capability is crucial for maintaining accurate mapping and localization over large areas. Given that VPR methods need to operate in real-time on embedded systems, it is critical to optimize these systems for minimal resource consumption. While the most efficient VPR approaches employ standard convolutional backbones with fixed descriptor dimensions, these often lead to redundancy in the embedding space as well as in the network architecture. Our work introduces a novel structured pruning method, to not only streamline common VPR architectures but also to strategically remove redundancies within the feature embedding space. This dual focus significantly enhances the efficiency of the system, reducing both map and model memory requirements and decreasing feature extraction and retrieval latencies. Our approach has reduced memory usage and latency by 21% and 16%, respectively, across models, while minimally impacting recall@1 accuracy by less than 1%. This significant improvement enhances real-time applications on edge devices with negligible accuracy loss.},
keywords = {accuracy, computational modelling, convolutional neural network, feature extraction, Localization, memory management, Real-time systems, representation learning, robustness, Vision-Based Navigation, visual place recognition (VPR), visualization},
pubstate = {published},
tppubtype = {article}
}
Visual Place Recognition (VPR) is fundamental for the global re-localization of robots and devices, enabling them to recognize previously visited locations based on visual inputs. This capability is crucial for maintaining accurate mapping and localization over large areas. Given that VPR methods need to operate in real-time on embedded systems, it is critical to optimize these systems for minimal resource consumption. While the most efficient VPR approaches employ standard convolutional backbones with fixed descriptor dimensions, these often lead to redundancy in the embedding space as well as in the network architecture. Our work introduces a novel structured pruning method, to not only streamline common VPR architectures but also to strategically remove redundancies within the feature embedding space. This dual focus significantly enhances the efficiency of the system, reducing both map and model memory requirements and decreasing feature extraction and retrieval latencies. Our approach has reduced memory usage and latency by 21% and 16%, respectively, across models, while minimally impacting recall@1 accuracy by less than 1%. This significant improvement enhances real-time applications on edge devices with negligible accuracy loss.
Kelly, Thomas Graham
Self-organised communication-aware control structures for robot swarms PhD Thesis
University of Southampton, 2025.
Abstract | Links | BibTeX | Tags:
@phdthesis{soton502200,
title = {Self-organised communication-aware control structures for robot swarms},
author = {Thomas Graham Kelly},
url = {https://eprints.soton.ac.uk/502200/},
year = {2025},
date = {2025-01-01},
publisher = {University of Southampton},
school = {University of Southampton},
abstract = {Robotic swarms are complex systems that rely on local communication between individual members of the swarm to spread information about their state and the environment. This information informs decisions and aids in tasks such as exploration and mapping. When communications break down between members of a swarm, it can become difficult to maintain accurate and up-to-date information about the state of the swarm and the environment. This problem is pertinent when humans are involved and may act as operators or teammates of the swarm. Here it is vital that the swarm can coordinate to distil and disseminate the vast amounts of information collected to the humans and throughout the swarm effectively, to maintain situational awareness.ensuremath<br/ensuremath>ensuremath<br/ensuremath>The collective decision-making of a swarm is one aspect that relies heavily on the ability to share information and observations to reach a swarm-wide consensus. This thesis investigates how communication constraints affect the swarm?s ability to reach a consensus and implement a communication-aware coordination strategy to mitigate these effects. We propose the communication-constrained collective decision-making problem and compare the performance of several collective decision-making strategies, enhanced with our coordination algorithm. We find that using such an approach improves the speed of a swarm to reach a consensus.ensuremath<br/ensuremath>ensuremath<br/ensuremath>Following on from this work, we examine a hybrid swarm system in a communication-limited environment. While swarms are traditionally considered decentralised systems, recent approaches have integrated decentralised and centralised control into a swarm system. We study the trade-offs in performance and communication in a hybrid system that can vary its control structure. We find that a higher level of centralisation does not guarantee higher performance and study how communication with a human operator is affected by the control structure. This work is extended to assess the feasibility of enabling a swarm system to learn the optimal control structure on the fly, according to mission requirements.},
keywords = {},
pubstate = {published},
tppubtype = {phdthesis}
}
Robotic swarms are complex systems that rely on local communication between individual members of the swarm to spread information about their state and the environment. This information informs decisions and aids in tasks such as exploration and mapping. When communications break down between members of a swarm, it can become difficult to maintain accurate and up-to-date information about the state of the swarm and the environment. This problem is pertinent when humans are involved and may act as operators or teammates of the swarm. Here it is vital that the swarm can coordinate to distil and disseminate the vast amounts of information collected to the humans and throughout the swarm effectively, to maintain situational awareness.ensuremath<br/ensuremath>ensuremath<br/ensuremath>The collective decision-making of a swarm is one aspect that relies heavily on the ability to share information and observations to reach a swarm-wide consensus. This thesis investigates how communication constraints affect the swarm?s ability to reach a consensus and implement a communication-aware coordination strategy to mitigate these effects. We propose the communication-constrained collective decision-making problem and compare the performance of several collective decision-making strategies, enhanced with our coordination algorithm. We find that using such an approach improves the speed of a swarm to reach a consensus.ensuremath<br/ensuremath>ensuremath<br/ensuremath>Following on from this work, we examine a hybrid swarm system in a communication-limited environment. While swarms are traditionally considered decentralised systems, recent approaches have integrated decentralised and centralised control into a swarm system. We study the trade-offs in performance and communication in a hybrid system that can vary its control structure. We find that a higher level of centralisation does not guarantee higher performance and study how communication with a human operator is affected by the control structure. This work is extended to assess the feasibility of enabling a swarm system to learn the optimal control structure on the fly, according to mission requirements.
Kiden, Sarah; Parisio, Isabela; Neff, Gina; Ramchurn, Gopal
AI management essentials (AIME) consultation response Technical Report
no. 10.5258/SOTON/PP0116, 2025.
Abstract | Links | BibTeX | Tags:
@techreport{soton501075,
title = {AI management essentials (AIME) consultation response},
author = {Sarah Kiden and Isabela Parisio and Gina Neff and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/501075/},
year = {2025},
date = {2025-01-01},
number = {10.5258/SOTON/PP0116},
publisher = {University of Southampton},
abstract = {We are submitting this response to the Department for Science, Innovation and Technology (DSIT) consultation on the new AI Management Essentials (AIME) tool1 on behalf of Responsible AI UK (RAi UK), an open and multidisciplinary network that brings together researchers from across the four nations of the UK to understand how we should shape the development of AI to benefit people, communities and society. To arrive at this response, we sent out a call to Principal Investigators and Co-Investigators of RAi UK funded projects2 to contribute to this consultation. What follows is a synthesis of the responses we received from our research community. Overall, RAi UK sees the AIME tool as a valuable first step for Small to Medium Sized Enterprises (SMEs) and Startups to adopt towards implementing robust and responsible AI governance practices. Our contributions below aim to improve the tool?s usability and impact.},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
We are submitting this response to the Department for Science, Innovation and Technology (DSIT) consultation on the new AI Management Essentials (AIME) tool1 on behalf of Responsible AI UK (RAi UK), an open and multidisciplinary network that brings together researchers from across the four nations of the UK to understand how we should shape the development of AI to benefit people, communities and society. To arrive at this response, we sent out a call to Principal Investigators and Co-Investigators of RAi UK funded projects2 to contribute to this consultation. What follows is a synthesis of the responses we received from our research community. Overall, RAi UK sees the AIME tool as a valuable first step for Small to Medium Sized Enterprises (SMEs) and Startups to adopt towards implementing robust and responsible AI governance practices. Our contributions below aim to improve the tool?s usability and impact.
Kiden, Sarah; Yazdanpanah, Vahid; Mestre, Rafael; Iusmen, Ingi; Atkinson, Joe; Parisio, Isabela; Ramchurn, Gopal; Stein, Sebastian
Human Rights and the Regulation of AI Written Evidence Technical Report
no. 10.5258/SOTON/PP0152, 2025.
@techreport{soton506667,
title = {Human Rights and the Regulation of AI Written Evidence},
author = {Sarah Kiden and Vahid Yazdanpanah and Rafael Mestre and Ingi Iusmen and Joe Atkinson and Isabela Parisio and Gopal Ramchurn and Sebastian Stein},
url = {https://eprints.soton.ac.uk/506667/},
year = {2025},
date = {2025-01-01},
number = {10.5258/SOTON/PP0152},
publisher = {University of Southampton},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
2024
Thavanesan, Navamayooran; Farahi, Arya; Parfitt, Charlotte; Belkhatir, Zehor; Azim, Tayyaba; Vallejos, Elvira Perez; Walters, Zoë; Ramchurn, Sarvapali; Underwood, Timothy J.; Vigneswaran, Ganesh
Insights from explainable AI in oesophageal cancer team decisions Journal Article
In: Computers in Biology and Medicine, vol. 180, 2024, (For the purpose of open access, the authors have applied a Creative Commons attribution license (CC-BY) to any Author Accepted Manuscript version arising from this submission.).
Abstract | Links | BibTeX | Tags: Decision-making, machine learning, Multidisciplinary teams, Oesophageal cancer
@article{soton493238,
title = {Insights from explainable AI in oesophageal cancer team decisions},
author = {Navamayooran Thavanesan and Arya Farahi and Charlotte Parfitt and Zehor Belkhatir and Tayyaba Azim and Elvira Perez Vallejos and Zoë Walters and Sarvapali Ramchurn and Timothy J. Underwood and Ganesh Vigneswaran},
url = {https://eprints.soton.ac.uk/493238/},
year = {2024},
date = {2024-08-01},
journal = {Computers in Biology and Medicine},
volume = {180},
abstract = {ensuremath<pensuremath>Background: clinician-led quality control into oncological decision-making is crucial for optimising patient care. Explainable artificial intelligence (XAI) techniques provide data-driven approaches to unravel how clinical variables influence this decision-making. We applied global XAI techniques to examine the impact of key clinical decision-drivers when mapped by a machine learning (ML) model, on the likelihood of receiving different oesophageal cancer (OC) treatment modalities by the multidisciplinary team (MDT).ensuremath</pensuremath>ensuremath<pensuremath>Methods: retrospective analysis of 893 OC patients managed between 2010 and 2022 at our tertiary unit, used a random forests (RF) classifier to predict four possible treatment pathways as determined by the MDT: neoadjuvant chemotherapy followed by surgery (NACT + S), neoadjuvant chemoradiotherapy followed by surgery (NACRT + S), surgery-alone, and palliative management. Variable importance and partial dependence (PD) analyses then examined the influence of targeted high-ranking clinical variables within the ML model on treatment decisions as a surrogate model of the MDT decision-making dynamic.�ensuremath</pensuremath>ensuremath<pensuremath>Results: amongst guideline-variables known to determine treatments, such as Tumour-Node-Metastasis (TNM) staging, age also proved highly important to the RF model (16.1 % of total importance) on variable importance analysis. PD subsequently revealed that predicted probabilities for all treatment modalities change significantly after 75 years (p < 0.001). Likelihood of surgery-alone and palliative therapies increased for patients aged 75?85yrs but lowered for NACT/NACRT. Performance status divided patients into two clusters which influenced all predicted outcomes in conjunction with age.�ensuremath</pensuremath>ensuremath<pensuremath>Conclusion: XAI techniques delineate the relationship between clinical factors and OC treatment decisions. These techniques identify advanced age as heavily influencing decisions based on our model with a greater role in patients with specific tumour characteristics. This study methodology provides the means for exploring conscious/subconscious bias and interrogating inconsistencies in team-based decision-making within the era of AI-driven decision support.ensuremath</pensuremath>},
note = {For the purpose of open access, the authors have applied a Creative Commons attribution license (CC-BY) to any Author Accepted Manuscript version arising from this submission.},
keywords = {Decision-making, machine learning, Multidisciplinary teams, Oesophageal cancer},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Background: clinician-led quality control into oncological decision-making is crucial for optimising patient care. Explainable artificial intelligence (XAI) techniques provide data-driven approaches to unravel how clinical variables influence this decision-making. We applied global XAI techniques to examine the impact of key clinical decision-drivers when mapped by a machine learning (ML) model, on the likelihood of receiving different oesophageal cancer (OC) treatment modalities by the multidisciplinary team (MDT).ensuremath</pensuremath>ensuremath<pensuremath>Methods: retrospective analysis of 893 OC patients managed between 2010 and 2022 at our tertiary unit, used a random forests (RF) classifier to predict four possible treatment pathways as determined by the MDT: neoadjuvant chemotherapy followed by surgery (NACT + S), neoadjuvant chemoradiotherapy followed by surgery (NACRT + S), surgery-alone, and palliative management. Variable importance and partial dependence (PD) analyses then examined the influence of targeted high-ranking clinical variables within the ML model on treatment decisions as a surrogate model of the MDT decision-making dynamic.�ensuremath</pensuremath>ensuremath<pensuremath>Results: amongst guideline-variables known to determine treatments, such as Tumour-Node-Metastasis (TNM) staging, age also proved highly important to the RF model (16.1 % of total importance) on variable importance analysis. PD subsequently revealed that predicted probabilities for all treatment modalities change significantly after 75 years (p < 0.001). Likelihood of surgery-alone and palliative therapies increased for patients aged 75?85yrs but lowered for NACT/NACRT. Performance status divided patients into two clusters which influenced all predicted outcomes in conjunction with age.�ensuremath</pensuremath>ensuremath<pensuremath>Conclusion: XAI techniques delineate the relationship between clinical factors and OC treatment decisions. These techniques identify advanced age as heavily influencing decisions based on our model with a greater role in patients with specific tumour characteristics. This study methodology provides the means for exploring conscious/subconscious bias and interrogating inconsistencies in team-based decision-making within the era of AI-driven decision support.ensuremath</pensuremath>
Naiseh, Mohammad; Webb, Catherine; Underwood, Tim; Ramchurn, Gopal; Walters, Zoe; Thavanesan, Navamayooran; Vigneswaran, Ganesh
XAI for group-AI interaction: towards collaborative and inclusive explanations Proceedings Article
In: Longo, Luca; Liu, Weiru; Montavon, Gregoire (Ed.): Joint Proceedings of the xAI 2024 Late-breaking Work, Demos and Doctoral Consortium co-located with the 2nd World Conference on eXplainable Artificial Intelligence (xAI 2024), pp. 249–256, CEUR Workshop Proceedings, 2024.
Abstract | Links | BibTeX | Tags: Explainable AI, Group-AI Interaction, Interaction Design
@inproceedings{soton497829,
title = {XAI for group-AI interaction: towards collaborative and inclusive explanations},
author = {Mohammad Naiseh and Catherine Webb and Tim Underwood and Gopal Ramchurn and Zoe Walters and Navamayooran Thavanesan and Ganesh Vigneswaran},
editor = {Luca Longo and Weiru Liu and Gregoire Montavon},
url = {https://eprints.soton.ac.uk/497829/},
year = {2024},
date = {2024-07-01},
booktitle = {Joint Proceedings of the xAI 2024 Late-breaking Work, Demos and Doctoral Consortium co-located with the 2nd World Conference on eXplainable Artificial Intelligence (xAI 2024)},
volume = {3793},
pages = {249–256},
publisher = {CEUR Workshop Proceedings},
abstract = {ensuremath<pensuremath>The increasing integration of Machine Learning (ML) into decision-making across various sectors has raised concerns about ethics, legality, explainability, and safety, highlighting the necessity of human oversight. In response, eXplainable AI (XAI) has emerged as a means to enhance transparency by providing insights into ML model decisions and offering humans an understanding of the underlying logic. Despite its potential, existing XAI models often lack practical usability and fail to improve human-AI performance, as they may introduce issues such as overreliance. This underscores the need for further research in Human-Centered XAI to improve the usability of current XAI methods. Notably, much of the current research focuses on one-to-one interactions between the XAI and individual decision-makers, overlooking the dynamics of many-to-one relationships in real-world scenarios where groups of humans collaborate using XAI in collective decision-making. In this late-breaking work, we draw upon current work in Human-Centered XAI research and discuss how XAI design could be transitioned to group-AI interaction. We discuss four potential challenges in the transition of XAI from human-AI interaction to group-AI interaction. This paper contributes to advancing the field of Human-Centered XAI and facilitates the discussion on group-XAI interaction, calling for further research in this area.ensuremath</pensuremath>},
keywords = {Explainable AI, Group-AI Interaction, Interaction Design},
pubstate = {published},
tppubtype = {inproceedings}
}
ensuremath<pensuremath>The increasing integration of Machine Learning (ML) into decision-making across various sectors has raised concerns about ethics, legality, explainability, and safety, highlighting the necessity of human oversight. In response, eXplainable AI (XAI) has emerged as a means to enhance transparency by providing insights into ML model decisions and offering humans an understanding of the underlying logic. Despite its potential, existing XAI models often lack practical usability and fail to improve human-AI performance, as they may introduce issues such as overreliance. This underscores the need for further research in Human-Centered XAI to improve the usability of current XAI methods. Notably, much of the current research focuses on one-to-one interactions between the XAI and individual decision-makers, overlooking the dynamics of many-to-one relationships in real-world scenarios where groups of humans collaborate using XAI in collective decision-making. In this late-breaking work, we draw upon current work in Human-Centered XAI research and discuss how XAI design could be transitioned to group-AI interaction. We discuss four potential challenges in the transition of XAI from human-AI interaction to group-AI interaction. This paper contributes to advancing the field of Human-Centered XAI and facilitates the discussion on group-XAI interaction, calling for further research in this area.ensuremath</pensuremath>
Early, Joseph Arthur
Interpretable multiple instance learning PhD Thesis
University of Southampton, 2024.
Abstract | Links | BibTeX | Tags:
@phdthesis{soton490767,
title = {Interpretable multiple instance learning},
author = {Joseph Arthur Early},
url = {https://eprints.soton.ac.uk/490767/},
year = {2024},
date = {2024-06-01},
publisher = {University of Southampton},
school = {University of Southampton},
abstract = {With the rising use of Artificial Intelligence (AI) and Machine Learning (ML) methods, there comes an increasing need to understand how automated systems make decisions. Interpretable ML provides insight into the underlying reasoning behind AI and ML models while not stifling their predictive performance. Doing so is important for many reasons, such as facilitating trust, increasing transparency, and providing improved collaboration and control through a better understanding of automated decision-making. Interpretability is very relevant across many ML paradigms and application domains. Multiple Instance Learning (MIL) is an ML paradigm where data are grouped into bags of instances, and only the bags are labelled (rather than each instance). This is beneficial in alleviating expensive labelling procedures and can be used to exploit the underlying structure of data. This thesis investigates how interpretability can be achieved within MIL. It begins with a formalisation of interpretable MIL, and then proposes a suite of model-agnostic post-hoc methods. This work is then extended to the specific application domain of high-resolution satellite imagery, using novel inherently interpretable MIL approaches that operate at multiple resolutions. Following on from work in the vision domain, new methods for interpretable MIL are developed for sequential data. First, it is explored in the domain of Reward Modelling (RM) for Reinforcement Learning (RL), demonstrating that interpretable MIL can be used to not only understand a model but also improve its predictive performance. This is mirrored in the application of interpretable MIL to Time Series Classification (TSC), where it is integrated into state-of-the-art methods and is able to improve both their interpretability and predictive performance. The integration into existing models to provide inherent interpretability means these benefits are delivered with little additional computational cost. ensuremath<br/ensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {phdthesis}
}
With the rising use of Artificial Intelligence (AI) and Machine Learning (ML) methods, there comes an increasing need to understand how automated systems make decisions. Interpretable ML provides insight into the underlying reasoning behind AI and ML models while not stifling their predictive performance. Doing so is important for many reasons, such as facilitating trust, increasing transparency, and providing improved collaboration and control through a better understanding of automated decision-making. Interpretability is very relevant across many ML paradigms and application domains. Multiple Instance Learning (MIL) is an ML paradigm where data are grouped into bags of instances, and only the bags are labelled (rather than each instance). This is beneficial in alleviating expensive labelling procedures and can be used to exploit the underlying structure of data. This thesis investigates how interpretability can be achieved within MIL. It begins with a formalisation of interpretable MIL, and then proposes a suite of model-agnostic post-hoc methods. This work is then extended to the specific application domain of high-resolution satellite imagery, using novel inherently interpretable MIL approaches that operate at multiple resolutions. Following on from work in the vision domain, new methods for interpretable MIL are developed for sequential data. First, it is explored in the domain of Reward Modelling (RM) for Reinforcement Learning (RL), demonstrating that interpretable MIL can be used to not only understand a model but also improve its predictive performance. This is mirrored in the application of interpretable MIL to Time Series Classification (TSC), where it is integrated into state-of-the-art methods and is able to improve both their interpretability and predictive performance. The integration into existing models to provide inherent interpretability means these benefits are delivered with little additional computational cost. ensuremath<br/ensuremath>
Kiden, Sarah; Stahl, Bernd; Townsend, Beverley; Maple, Carsten; Vincent, Charles; Sampson, Fraser; Gilbert, Geoff; Smith, Helen; Deshmukh, Jayati; Ross, Jen; Williams, Jennifer; Rincon, Jesus Martinez; Lisinska, Justyna; O?Shea, Karen; Abreu, Márjory Da Costa; Bencomo, Nelly; Deb, Oishi; Winter, Peter; Li, Phoebe; Torr, Philip; Lau, Pin Lean; Iniesta, Raquel; Ramchurn, Gopal; Stein, Sebastian; Yazdanpanah, Vahid
Responsible AI governance: A response to UN interim report on governing AI for humanity Technical Report
no. 10.5258/SOTON/PP0057, 2024.
@techreport{soton488908,
title = {Responsible AI governance: A response to UN interim report on governing AI for humanity},
author = {Sarah Kiden and Bernd Stahl and Beverley Townsend and Carsten Maple and Charles Vincent and Fraser Sampson and Geoff Gilbert and Helen Smith and Jayati Deshmukh and Jen Ross and Jennifer Williams and Jesus Martinez Rincon and Justyna Lisinska and Karen O?Shea and Márjory Da Costa Abreu and Nelly Bencomo and Oishi Deb and Peter Winter and Phoebe Li and Philip Torr and Pin Lean Lau and Raquel Iniesta and Gopal Ramchurn and Sebastian Stein and Vahid Yazdanpanah},
url = {https://eprints.soton.ac.uk/488908/},
year = {2024},
date = {2024-03-01},
number = {10.5258/SOTON/PP0057},
publisher = {Public Policy, University of Southampton},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
Abioye, Ayodeji O.; Hunt, William; Gu, Yue; Schneiders, Eike; Naiseh, Mohammad; Fischer, Joel E.; Ramchurn, Sarvapali D.; Soorati, Mohammad D.; Archibald, Blair; Sevegnani, Michele
The effect of predictive formal modelling at runtime on performance in human-swarm interaction Proceedings Article
In: HRI '24: Companion of the 2024 ACM/IEEE International Conference on Human-Robot Interaction, pp. 172?176, Association for Computing Machinery, 2024, (Publisher Copyright: © 2024 Copyright held by the owner/author(s)).
Abstract | Links | BibTeX | Tags: Human-Robot Interaction (HRI), Human-Swarm Interaction (HSI), Predictive Formal Modelling (PFM), Task Performance
@inproceedings{soton488273,
title = {The effect of predictive formal modelling at runtime on performance in human-swarm interaction},
author = {Ayodeji O. Abioye and William Hunt and Yue Gu and Eike Schneiders and Mohammad Naiseh and Joel E. Fischer and Sarvapali D. Ramchurn and Mohammad D. Soorati and Blair Archibald and Michele Sevegnani},
url = {https://eprints.soton.ac.uk/488273/},
year = {2024},
date = {2024-03-01},
booktitle = {HRI '24: Companion of the 2024 ACM/IEEE International Conference on Human-Robot Interaction},
pages = {172?176},
publisher = {Association for Computing Machinery},
abstract = {Formal Modelling is often used as part of the design and testing process of software development to ensure that components operate within suitable bounds even in unexpected circumstances. In this paper, we use predictive formal modelling (PFM) at runtime in a human-swarm mission and show that this integration can be used to improve the performance of human-swarm teams. We recruited 60 participants to operate a simulated aerial swarm to deliver parcels to target locations. In the PFM condition, operators were informed of the estimated completion times given the number of drones deployed, whereas, in the No-PFM condition, operators did not have this information. The operators could control the mission by adding or removing drones from the mission and thereby, increasing or decreasing the overall mission cost. The evaluation of human-swarm performance relied on four metrics: the task completion time, the number of agents, the number of completed tasks, and the cost per task. Our results show that PFM modelling at runtime improves mission performance without significantly affecting the operator's workload or the system's usability.},
note = {Publisher Copyright:
© 2024 Copyright held by the owner/author(s)},
keywords = {Human-Robot Interaction (HRI), Human-Swarm Interaction (HSI), Predictive Formal Modelling (PFM), Task Performance},
pubstate = {published},
tppubtype = {inproceedings}
}
Formal Modelling is often used as part of the design and testing process of software development to ensure that components operate within suitable bounds even in unexpected circumstances. In this paper, we use predictive formal modelling (PFM) at runtime in a human-swarm mission and show that this integration can be used to improve the performance of human-swarm teams. We recruited 60 participants to operate a simulated aerial swarm to deliver parcels to target locations. In the PFM condition, operators were informed of the estimated completion times given the number of drones deployed, whereas, in the No-PFM condition, operators did not have this information. The operators could control the mission by adding or removing drones from the mission and thereby, increasing or decreasing the overall mission cost. The evaluation of human-swarm performance relied on four metrics: the task completion time, the number of agents, the number of completed tasks, and the cost per task. Our results show that PFM modelling at runtime improves mission performance without significantly affecting the operator's workload or the system's usability.
Soorati, Mohammad D.; Naiseh, Mohammad; Hunt, William; Parnell, Katie; Clark, Jediah; Ramchurn, Sarvapali D.
Enabling trustworthiness in human-swarm systems through a digital twin Book Section
In: Dasgupta, Prithviraj; Llinas, James; Gillespie, Tony; Fouse, Scott; Lawless, William; Mittu, Ranjeev; Sofge, Donlad (Ed.): Putting AI in the Critical Loop: Assured Trust and Autonomy in Human-Machine Teams, pp. 93–125, Academic Press, 2024, (Publisher Copyright: © 2024 Elsevier Inc. All rights reserved.).
Abstract | Links | BibTeX | Tags: Digital twin, Explainability, Human-swarm interaction, Trustworthy Autonomous Systems, User-centered design
@incollection{soton491769,
title = {Enabling trustworthiness in human-swarm systems through a digital twin},
author = {Mohammad D. Soorati and Mohammad Naiseh and William Hunt and Katie Parnell and Jediah Clark and Sarvapali D. Ramchurn},
editor = {Prithviraj Dasgupta and James Llinas and Tony Gillespie and Scott Fouse and William Lawless and Ranjeev Mittu and Donlad Sofge},
url = {https://eprints.soton.ac.uk/491769/},
year = {2024},
date = {2024-02-01},
booktitle = {Putting AI in the Critical Loop: Assured Trust and Autonomy in Human-Machine Teams},
pages = {93–125},
publisher = {Academic Press},
abstract = {Robot swarms are highly dynamic systems that exhibit fault-tolerant behavior in accomplishing given tasks. Applications of swarm robotics are very limited due to the lack of complex decision-making capability. Real-world applications are only possible if we use human supervision to monitor and control the behavior of the swarm. Ensuring that human operators can trust the swarm system is one of the key challenges in human-swarm systems. This chapter presents a digital twin for trustworthy human-swarm teaming. The first element in designing such a simulation platform is to understand the trust requirements to label a human-swarm system as trustworthy. In order to outline the key trust requirements, we interviewed a group of experienced uncrewed aerial vehicle (UAV) operators and collated their suggestions for building and repairing trusts in single and multiple UAV systems. We then performed a survey to gather swarm experts? points of view on creating a taxonomy for explainability in human-swarm systems. This chapter presents a digital twin platform that implements a disaster management use case and has the capacity to meet the extracted trust and explainability requirements.},
note = {Publisher Copyright:
© 2024 Elsevier Inc. All rights reserved.},
keywords = {Digital twin, Explainability, Human-swarm interaction, Trustworthy Autonomous Systems, User-centered design},
pubstate = {published},
tppubtype = {incollection}
}
Robot swarms are highly dynamic systems that exhibit fault-tolerant behavior in accomplishing given tasks. Applications of swarm robotics are very limited due to the lack of complex decision-making capability. Real-world applications are only possible if we use human supervision to monitor and control the behavior of the swarm. Ensuring that human operators can trust the swarm system is one of the key challenges in human-swarm systems. This chapter presents a digital twin for trustworthy human-swarm teaming. The first element in designing such a simulation platform is to understand the trust requirements to label a human-swarm system as trustworthy. In order to outline the key trust requirements, we interviewed a group of experienced uncrewed aerial vehicle (UAV) operators and collated their suggestions for building and repairing trusts in single and multiple UAV systems. We then performed a survey to gather swarm experts? points of view on creating a taxonomy for explainability in human-swarm systems. This chapter presents a digital twin platform that implements a disaster management use case and has the capacity to meet the extracted trust and explainability requirements.
Thavanesan, Navamayooran; Parfitt, Charlotte; Bodala, Indu; Walters, Zoë; Ramchurn, Sarvapali; Underwood, Timothy; Vigneswaran, Ganesh
Machine learning models for curative and palliative oesophageal cancer treatment pathway prediction Miscellaneous
2024.
Abstract | Links | BibTeX | Tags:
@misc{soton497828,
title = {Machine learning models for curative and palliative oesophageal cancer treatment pathway prediction},
author = {Navamayooran Thavanesan and Charlotte Parfitt and Indu Bodala and Zoë Walters and Sarvapali Ramchurn and Timothy Underwood and Ganesh Vigneswaran},
url = {https://eprints.soton.ac.uk/497828/},
year = {2024},
date = {2024-01-01},
journal = {European Journal of Surgical Oncology},
volume = {50},
number = {1},
abstract = {Introduction: Oesophageal Cancer Multidisciplinary Teams (OC MDTs) operate under significant caseload pressures. This risks variability of decision-making which may influence patient outcomes. Machine Learning (ML) offers the ability to streamline and standardise decision-making by learning from historic treatment decisions to prediction treatment for new patients. We present internally validated ML models designed to predict OC MDT treatment decisions for curative and palliative OC patients.ensuremath<br/ensuremath>ensuremath<br/ensuremath>Methods: four ML algorithms (multinomial logistic regression (MLR), random forests (RF), extreme gradient boost (XGB) and decision tree (DT)) were trained using nested cross-validation on a cohort of 938 OC cases from a single tertiary unit over a 12-year period. The models classified predicted treatments into one of: Surgery (S), Neoadjuvant Chemotherapy (NACT) + S, Neoadjuvant Chemoradiotherapy (NACRT) + S, Endoscopic or Palliative treatment. Performance was assessed on Area Under the Curve (AUC).ensuremath<br/ensuremath>ensuremath<br/ensuremath>Results: across algorithms, all models performed strongly with mean AUC for Surgery = 0.849$±$0.026, NACT +S = 0.884$±$0.008, NACRT +S = 0.834$±$0.035},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Introduction: Oesophageal Cancer Multidisciplinary Teams (OC MDTs) operate under significant caseload pressures. This risks variability of decision-making which may influence patient outcomes. Machine Learning (ML) offers the ability to streamline and standardise decision-making by learning from historic treatment decisions to prediction treatment for new patients. We present internally validated ML models designed to predict OC MDT treatment decisions for curative and palliative OC patients.ensuremath<br/ensuremath>ensuremath<br/ensuremath>Methods: four ML algorithms (multinomial logistic regression (MLR), random forests (RF), extreme gradient boost (XGB) and decision tree (DT)) were trained using nested cross-validation on a cohort of 938 OC cases from a single tertiary unit over a 12-year period. The models classified predicted treatments into one of: Surgery (S), Neoadjuvant Chemotherapy (NACT) + S, Neoadjuvant Chemoradiotherapy (NACRT) + S, Endoscopic or Palliative treatment. Performance was assessed on Area Under the Curve (AUC).ensuremath<br/ensuremath>ensuremath<br/ensuremath>Results: across algorithms, all models performed strongly with mean AUC for Surgery = 0.849$±$0.026, NACT +S = 0.884$±$0.008, NACRT +S = 0.834$±$0.035
Kelly, Thomas Graham; Soorati, Mohammad; Zauner, Klaus-Peter; Ramchurn, Gopal; Tarapore, Danesh
Trade-offs of dynamic control structure in human-swarm systems Proceedings Article
In: The International Symposium on Distributed Autonomous Robotic Systems (DARS) 2024, 2024.
Abstract | Links | BibTeX | Tags:
@inproceedings{soton492838,
title = {Trade-offs of dynamic control structure in human-swarm systems},
author = {Thomas Graham Kelly and Mohammad Soorati and Klaus-Peter Zauner and Gopal Ramchurn and Danesh Tarapore},
url = {https://eprints.soton.ac.uk/492838/},
year = {2024},
date = {2024-01-01},
booktitle = {The International Symposium on Distributed Autonomous Robotic Systems (DARS) 2024},
abstract = {Swarm robotics is a study of simple robots that exhibit complex behaviour only by interacting locally with other robots and their environment. The control in swarm robotics is mainly distributed whereas centralised control is widely used in other fields of robotics. Centralised and decentralised control strategies both pose a unique set of benefits and drawbacks for the control of multi-robot systems. While decentralised systems are more scalable and resilient, they are less efficient compared to the centralised systems and they lead to excessive data transmissions to the human operators causing cognitive overload. We examine the trade-offs of each of these approaches in a human-swarm system to perform an environmental monitoring task and propose a flexible hybrid approach, which combines elements of hierarchical and decentralised systems. We find that a flexible hybrid system can outperform a centralised system (in our environmental monitoring task by 19.2%) while reducing the number of messages sent to a human operator (here by 23.1%). We conclude that establishing centralisation for a system is not always optimal for performance and that utilising aspects of centralised and decentralised systems can keep the swarm from hindering its performance.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Swarm robotics is a study of simple robots that exhibit complex behaviour only by interacting locally with other robots and their environment. The control in swarm robotics is mainly distributed whereas centralised control is widely used in other fields of robotics. Centralised and decentralised control strategies both pose a unique set of benefits and drawbacks for the control of multi-robot systems. While decentralised systems are more scalable and resilient, they are less efficient compared to the centralised systems and they lead to excessive data transmissions to the human operators causing cognitive overload. We examine the trade-offs of each of these approaches in a human-swarm system to perform an environmental monitoring task and propose a flexible hybrid approach, which combines elements of hierarchical and decentralised systems. We find that a flexible hybrid system can outperform a centralised system (in our environmental monitoring task by 19.2%) while reducing the number of messages sent to a human operator (here by 23.1%). We conclude that establishing centralisation for a system is not always optimal for performance and that utilising aspects of centralised and decentralised systems can keep the swarm from hindering its performance.
2023
Early, Joseph; Deweese, Ying-Jung Chen; Evers, Christine; Ramchurn, Sarvapali
Extending scene-to-patch models: Multi-resolution multiple instance learning for Earth observation Journal Article
In: Environmental Data Science, vol. 2, pp. 18, 2023.
Abstract | Links | BibTeX | Tags:
@article{soton490766,
title = {Extending scene-to-patch models: Multi-resolution multiple instance learning for Earth observation},
author = {Joseph Early and Ying-Jung Chen Deweese and Christine Evers and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/490766/},
year = {2023},
date = {2023-12-01},
journal = {Environmental Data Science},
volume = {2},
pages = {18},
abstract = {Land cover classification (LCC) and natural disaster response (NDR) are important issues in climate change mitigation and adaptation. Existing approaches that use machine learning with Earth observation (EO) imaging data for LCC and NDR often rely on fully annotated and segmented datasets. Creating these datasets requires a large amount of effort, and a lack of suitable datasets has become an obstacle in scaling the use of machine learning for EO. In this study, we extend our prior work on Scene-to-Patch models: an alternative machine learning approach for EO that utilizes Multiple Instance Learning (MIL). As our approach only requires high-level scene labels, it enables much faster development of new datasets while still providing segmentation through patch-level predictions, ultimately increasing the accessibility of using machine learning for EO. We propose new multi-resolution MIL architectures that outperform single-resolution MIL models and non-MIL baselines on the DeepGlobe LCC and FloodNet NDR datasets. In addition, we conduct a thorough analysis of model performance and interpretability.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Land cover classification (LCC) and natural disaster response (NDR) are important issues in climate change mitigation and adaptation. Existing approaches that use machine learning with Earth observation (EO) imaging data for LCC and NDR often rely on fully annotated and segmented datasets. Creating these datasets requires a large amount of effort, and a lack of suitable datasets has become an obstacle in scaling the use of machine learning for EO. In this study, we extend our prior work on Scene-to-Patch models: an alternative machine learning approach for EO that utilizes Multiple Instance Learning (MIL). As our approach only requires high-level scene labels, it enables much faster development of new datasets while still providing segmentation through patch-level predictions, ultimately increasing the accessibility of using machine learning for EO. We propose new multi-resolution MIL architectures that outperform single-resolution MIL models and non-MIL baselines on the DeepGlobe LCC and FloodNet NDR datasets. In addition, we conduct a thorough analysis of model performance and interpretability.
Rigley, Eryn; Bentley, Caitlin; Krook, Joshua; Ramchurn, Gopal
Evaluating international AI skills policy: a systematic review of AI skills policy in seven countries Journal Article
In: Global Policy, 2023, (Funding Information: This research was supported via UKRI by the DCMS Science and Analysis R&D Programme. It was developed and produced according to UKRI's initial hypotheses and output requests. Any primary research, subsequent findings or recommendations do not represent Government views or policy and are produced according to academic ethics, quality assurance and independence.).
Abstract | Links | BibTeX | Tags:
@article{soton485727,
title = {Evaluating international AI skills policy: a systematic review of AI skills policy in seven countries},
author = {Eryn Rigley and Caitlin Bentley and Joshua Krook and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/485727/},
year = {2023},
date = {2023-12-01},
journal = {Global Policy},
abstract = {ensuremath<pensuremath>As artificial intelligence (AI) is having an increasingly disruptive impact across industries, companies continue to report having difficulty when recruiting for AI roles, while new graduates find it difficult to find employment, indicating a skills gap or skills misalignment. International approaches to AI skills programmes can offer a guide to future policy development of a skilled workforce, best placed to harness the economic opportunities that AI may support. The authors performed a systematic literature review on AI skills in government policies and documents from seven countries: Australia, Canada, China, Singapore, Sweden, the United Kingom and the United States. We found a divide between countries which emphasised a broader, nationwide approach to upskill and educate all citizens at different levels, namely the United States and Singapore and those countries which emphasised a narrower focus on educating a smaller group of experts with advanced AI knowledge and skills, namely China, Sweden and Canada. We found that the former, broader approaches tended to correlate with higher AI readiness and index scores than the narrower, expert-driven approach. Our findings indicate that, to match world-leading AI readiness, future AI skills policy should follow these broad, nationwide approaches to upskill and educate all citizens at different levels of AI expertise.ensuremath</pensuremath>},
note = {Funding Information:
This research was supported via UKRI by the DCMS Science and Analysis R&D Programme. It was developed and produced according to UKRI's initial hypotheses and output requests. Any primary research, subsequent findings or recommendations do not represent Government views or policy and are produced according to academic ethics, quality assurance and independence.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>As artificial intelligence (AI) is having an increasingly disruptive impact across industries, companies continue to report having difficulty when recruiting for AI roles, while new graduates find it difficult to find employment, indicating a skills gap or skills misalignment. International approaches to AI skills programmes can offer a guide to future policy development of a skilled workforce, best placed to harness the economic opportunities that AI may support. The authors performed a systematic literature review on AI skills in government policies and documents from seven countries: Australia, Canada, China, Singapore, Sweden, the United Kingom and the United States. We found a divide between countries which emphasised a broader, nationwide approach to upskill and educate all citizens at different levels, namely the United States and Singapore and those countries which emphasised a narrower focus on educating a smaller group of experts with advanced AI knowledge and skills, namely China, Sweden and Canada. We found that the former, broader approaches tended to correlate with higher AI readiness and index scores than the narrower, expert-driven approach. Our findings indicate that, to match world-leading AI readiness, future AI skills policy should follow these broad, nationwide approaches to upskill and educate all citizens at different levels of AI expertise.ensuremath</pensuremath>
Singh, Lokesh; Ramchurn, Gopal
The effect of automated agents on individual performance under induced stress Proceedings Article
In: Kalra, Jay (Ed.): Emerging Technologies in Healthcare and Medicine: Proceedings of the AHFE International Conference on Human Factors in Design, Engineering and Computing (AHFE 2023 Hawaii Edition), pp. 118–127, AHFE International, 2023.
Abstract | Links | BibTeX | Tags: Decision-making, Human-agent, Individual performance, Induced stress, Time pressure
@inproceedings{soton485655,
title = {The effect of automated agents on individual performance under induced stress},
author = {Lokesh Singh and Gopal Ramchurn},
editor = {Jay Kalra},
url = {https://eprints.soton.ac.uk/485655/},
year = {2023},
date = {2023-11-01},
booktitle = {Emerging Technologies in Healthcare and Medicine: Proceedings of the AHFE International Conference on Human Factors in Design, Engineering and Computing (AHFE 2023 Hawaii Edition)},
pages = {118–127},
publisher = {AHFE International},
abstract = {Induced stress is a phenomenon commonly experienced across different fields such as emergency services, healthcare, air traffic control, sports, and business - which necessitates the development of effective coping strategies and resilience for individuals or teams performing under pressure. This study aims to examine the effects of automated agents on individual performance during high-stress conditions. The design of these agents ensures they carry out identical tasks as participants based on predetermined frameworks. Participants underwent an experimentally designed task that aimed at inducing stress while measuring their performance amidst time pressure and auditory distraction. Results indicate that working with automated agents causes individuals to alter their approach by focusing narrowly on immediate concerns - making it challenging for them to consider several options or see broader contexts accurately. Regardless of ability level participants' performances were influenced by these automated agents. Future research will explore how these findings interact with physiological signals. This study highlights the importance of developing effective coping strategies and the potential impact of social factors on individual performance under induced stress.},
keywords = {Decision-making, Human-agent, Individual performance, Induced stress, Time pressure},
pubstate = {published},
tppubtype = {inproceedings}
}
Induced stress is a phenomenon commonly experienced across different fields such as emergency services, healthcare, air traffic control, sports, and business - which necessitates the development of effective coping strategies and resilience for individuals or teams performing under pressure. This study aims to examine the effects of automated agents on individual performance during high-stress conditions. The design of these agents ensures they carry out identical tasks as participants based on predetermined frameworks. Participants underwent an experimentally designed task that aimed at inducing stress while measuring their performance amidst time pressure and auditory distraction. Results indicate that working with automated agents causes individuals to alter their approach by focusing narrowly on immediate concerns - making it challenging for them to consider several options or see broader contexts accurately. Regardless of ability level participants' performances were influenced by these automated agents. Future research will explore how these findings interact with physiological signals. This study highlights the importance of developing effective coping strategies and the potential impact of social factors on individual performance under induced stress.
Thavanesan, Navamayooran; Bodala, Indu; Walters, Zoe; Ramchurn, Sarvapali; Underwood, Timothy J.; Vigneswaran, Ganesh
Machine learning to predict curative multidisciplinary team treatment decisions in oesophageal cancer Journal Article
In: European Journal of Surgical Oncology, vol. 49, no. 11, 2023, (Publisher Copyright: © 2023 The Author(s)).
Abstract | Links | BibTeX | Tags: Artificial Intelligence, machine learning, Oesophageal cancer multidisciplinary team
@article{soton479497b,
title = {Machine learning to predict curative multidisciplinary team treatment decisions in oesophageal cancer},
author = {Navamayooran Thavanesan and Indu Bodala and Zoe Walters and Sarvapali Ramchurn and Timothy J. Underwood and Ganesh Vigneswaran},
url = {https://eprints.soton.ac.uk/479497/},
year = {2023},
date = {2023-11-01},
journal = {European Journal of Surgical Oncology},
volume = {49},
number = {11},
abstract = {ensuremath<pensuremath>Background: Rising workflow pressures within the oesophageal cancer (OC) multidisciplinary team (MDT) can lead to variability in decision-making, and health inequality. Machine learning (ML) offers a potential automated data-driven approach to address inconsistency and standardize care. The aim of this experimental pilot study was to develop ML models able to predict curative OC MDT treatment decisions and determine the relative importance of underlying decision-critical variables. Methods: Retrospective complete-case analysis of oesophagectomy patients $±$ neoadjuvant chemotherapy (NACT) or chemoradiotherapy (NACRT) between 2010 and 2020. Established ML algorithms (Multinomial Logistic regression (MLR), Random Forests (RF), Extreme Gradient Boosting (XGB)) and Decision Tree (DT) were used to train models predicting OC MDT treatment decisions: surgery (S), NACT + S or NACRT + S. Performance metrics included Area Under the Curve (AUC), Accuracy, Kappa, LogLoss, F1 and Precision -Recall AUC. Variable importance was calculated for each model. Results: We identified 399 cases with a male-to-female ratio of 3.6:1 and median age of 66.1yrs (range 32?83). MLR outperformed RF, XGB and DT across performance metrics (mean AUC of 0.793 [$±$0.045] vs 0.757 [$±$0.068], 0.740 [$±$0.042], and 0.709 [$±$0.021] respectively). Variable importance analysis identified age as a major factor in the decision to offer surgery alone or NACT + S across models (p < 0.05). Conclusions: ML techniques can use limited feature-sets to predict curative UGI MDT treatment decisions. Explainable Artificial Intelligence methods provide insight into decision-critical variables, highlighting underlying subconscious biases in cancer care decision-making. Such models may allow prioritization of caseload, improve efficiency, and offer data-driven decision-assistance to MDTs in the future.ensuremath</pensuremath>},
note = {Publisher Copyright:
© 2023 The Author(s)},
keywords = {Artificial Intelligence, machine learning, Oesophageal cancer multidisciplinary team},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Background: Rising workflow pressures within the oesophageal cancer (OC) multidisciplinary team (MDT) can lead to variability in decision-making, and health inequality. Machine learning (ML) offers a potential automated data-driven approach to address inconsistency and standardize care. The aim of this experimental pilot study was to develop ML models able to predict curative OC MDT treatment decisions and determine the relative importance of underlying decision-critical variables. Methods: Retrospective complete-case analysis of oesophagectomy patients $±$ neoadjuvant chemotherapy (NACT) or chemoradiotherapy (NACRT) between 2010 and 2020. Established ML algorithms (Multinomial Logistic regression (MLR), Random Forests (RF), Extreme Gradient Boosting (XGB)) and Decision Tree (DT) were used to train models predicting OC MDT treatment decisions: surgery (S), NACT + S or NACRT + S. Performance metrics included Area Under the Curve (AUC), Accuracy, Kappa, LogLoss, F1 and Precision -Recall AUC. Variable importance was calculated for each model. Results: We identified 399 cases with a male-to-female ratio of 3.6:1 and median age of 66.1yrs (range 32?83). MLR outperformed RF, XGB and DT across performance metrics (mean AUC of 0.793 [$±$0.045] vs 0.757 [$±$0.068], 0.740 [$±$0.042], and 0.709 [$±$0.021] respectively). Variable importance analysis identified age as a major factor in the decision to offer surgery alone or NACT + S across models (p < 0.05). Conclusions: ML techniques can use limited feature-sets to predict curative UGI MDT treatment decisions. Explainable Artificial Intelligence methods provide insight into decision-critical variables, highlighting underlying subconscious biases in cancer care decision-making. Such models may allow prioritization of caseload, improve efficiency, and offer data-driven decision-assistance to MDTs in the future.ensuremath</pensuremath>
Krook, Joshua; Williams, Jennifer; Seabrooke, Tina; Schneiders, Eike; Blockx, Jan; Middleton, Stuart E; Ramchurn, Sarvapali
AI large language models inquiry: TASHub Response Miscellaneous
2023.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, chatgpt, european law, generative ai, Large Language Models, law and technology, technology policy
@misc{soton481740,
title = {AI large language models inquiry: TASHub Response},
author = {Joshua Krook and Jennifer Williams and Tina Seabrooke and Eike Schneiders and Jan Blockx and Stuart E Middleton and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/481740/},
year = {2023},
date = {2023-08-01},
publisher = {University of Southampton},
abstract = {Policy submission to the Consultation by Communications and Digital Committee, House of Lords, AI Large Language Models Inquiry.ensuremath<br/ensuremath>},
keywords = {Artificial Intelligence, chatgpt, european law, generative ai, Large Language Models, law and technology, technology policy},
pubstate = {published},
tppubtype = {misc}
}
Policy submission to the Consultation by Communications and Digital Committee, House of Lords, AI Large Language Models Inquiry.ensuremath<br/ensuremath>
Krook, Joshua; Williams, Jennifer; Seabrooke, Tina; Schneiders, Eike; Blockx, Jan; Middleton, Stuart E; Ramchurn, Sarvapali
AI large language models inquiry: TASHub response Miscellaneous
2023.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, chatgpt, european law, generative ai, Large Language Models, law and technology, technology policy
@misc{soton481740b,
title = {AI large language models inquiry: TASHub response},
author = {Joshua Krook and Jennifer Williams and Tina Seabrooke and Eike Schneiders and Jan Blockx and Stuart E Middleton and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/481740/},
year = {2023},
date = {2023-08-01},
publisher = {University of Southampton},
abstract = {Policy submission to the Consultation by Communications and Digital Committee, House of Lords, AI Large Language Models Inquiry.ensuremath<br/ensuremath>},
keywords = {Artificial Intelligence, chatgpt, european law, generative ai, Large Language Models, law and technology, technology policy},
pubstate = {published},
tppubtype = {misc}
}
Policy submission to the Consultation by Communications and Digital Committee, House of Lords, AI Large Language Models Inquiry.ensuremath<br/ensuremath>
Thavanesan, Navamayooran; Bodala, Indu; Walters, Zoe; Ramchurn, Sarvapali; Underwood, Timothy J; Vigneswaran, Ganesh
Machine learning to predict curative multidisciplinary team treatment decisions in oesophageal cancer Journal Article
In: European Journal of Surgical Oncology, 2023, (Publisher Copyright: copyright 2023 The Author(s)).
Abstract | Links | BibTeX | Tags: Artificial Intelligence, machine learning, Oesophageal cancer multidisciplinary team
@article{soton479497,
title = {Machine learning to predict curative multidisciplinary team treatment decisions in oesophageal cancer},
author = {Navamayooran Thavanesan and Indu Bodala and Zoe Walters and Sarvapali Ramchurn and Timothy J Underwood and Ganesh Vigneswaran},
url = {https://eprints.soton.ac.uk/479497/},
year = {2023},
date = {2023-07-01},
journal = {European Journal of Surgical Oncology},
abstract = {ensuremath<pensuremath>Background: Rising workflow pressures within the oesophageal cancer (OC) multidisciplinary team (MDT) can lead to variability in decision-making, and health inequality. Machine learning (ML) offers a potential automated data-driven approach to address inconsistency and standardize care. The aim of this experimental pilot study was to develop ML models able to predict curative OC MDT treatment decisions and determine the relative importance of underlying decision-critical variables. Methods: Retrospective complete-case analysis of oesophagectomy patients $pm$ neoadjuvant chemotherapy (NACT) or chemoradiotherapy (NACRT) between 2010 and 2020. Established ML algorithms (Multinomial Logistic regression (MLR), Random Forests (RF), Extreme Gradient Boosting (XGB)) and Decision Tree (DT) were used to train models predicting OC MDT treatment decisions: surgery (S), NACT + S or NACRT + S. Performance metrics included Area Under the Curve (AUC), Accuracy, Kappa, LogLoss, F1 and Precision -Recall AUC. Variable importance was calculated for each model. Results: We identified 399 cases with a male-to-female ratio of 3.6:1 and median age of 66.1yrs (range 32?83). MLR outperformed RF, XGB and DT across performance metrics (mean AUC of 0.793 [$pm$0.045] vs 0.757 [$pm$0.068], 0.740 [$pm$0.042], and 0.709 [$pm$0.021] respectively). Variable importance analysis identified age as a major factor in the decision to offer surgery alone or NACT + S across models (p < 0.05). Conclusions: ML techniques can use limited feature-sets to predict curative UGI MDT treatment decisions. Explainable Artificial Intelligence methods provide insight into decision-critical variables, highlighting underlying subconscious biases in cancer care decision-making. Such models may allow prioritization of caseload, improve efficiency, and offer data-driven decision-assistance to MDTs in the future.ensuremath</pensuremath>},
note = {Publisher Copyright:
copyright 2023 The Author(s)},
keywords = {Artificial Intelligence, machine learning, Oesophageal cancer multidisciplinary team},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Background: Rising workflow pressures within the oesophageal cancer (OC) multidisciplinary team (MDT) can lead to variability in decision-making, and health inequality. Machine learning (ML) offers a potential automated data-driven approach to address inconsistency and standardize care. The aim of this experimental pilot study was to develop ML models able to predict curative OC MDT treatment decisions and determine the relative importance of underlying decision-critical variables. Methods: Retrospective complete-case analysis of oesophagectomy patients $pm$ neoadjuvant chemotherapy (NACT) or chemoradiotherapy (NACRT) between 2010 and 2020. Established ML algorithms (Multinomial Logistic regression (MLR), Random Forests (RF), Extreme Gradient Boosting (XGB)) and Decision Tree (DT) were used to train models predicting OC MDT treatment decisions: surgery (S), NACT + S or NACRT + S. Performance metrics included Area Under the Curve (AUC), Accuracy, Kappa, LogLoss, F1 and Precision -Recall AUC. Variable importance was calculated for each model. Results: We identified 399 cases with a male-to-female ratio of 3.6:1 and median age of 66.1yrs (range 32?83). MLR outperformed RF, XGB and DT across performance metrics (mean AUC of 0.793 [$pm$0.045] vs 0.757 [$pm$0.068], 0.740 [$pm$0.042], and 0.709 [$pm$0.021] respectively). Variable importance analysis identified age as a major factor in the decision to offer surgery alone or NACT + S across models (p < 0.05). Conclusions: ML techniques can use limited feature-sets to predict curative UGI MDT treatment decisions. Explainable Artificial Intelligence methods provide insight into decision-critical variables, highlighting underlying subconscious biases in cancer care decision-making. Such models may allow prioritization of caseload, improve efficiency, and offer data-driven decision-assistance to MDTs in the future.ensuremath</pensuremath>
Abioye, Ayodeji
University of Southampton, 2023.
Abstract | Links | BibTeX | Tags:
@phdthesis{soton479472,
title = {Multimodal speech and visual gesture control interface technique for small unmanned multirotor aircraft},
author = {Ayodeji Abioye},
url = {https://eprints.soton.ac.uk/479472/},
year = {2023},
date = {2023-07-01},
publisher = {University of Southampton},
school = {University of Southampton},
abstract = {ensuremath<p class="MsoNormal"ensuremath>This research conducted an investigation into the use of novel human computer interaction(HCI) interfaces in the control of small multirotor unmanned aerial vehicles(UAVs). The main objective was to propose, design, and develop an alternative control interface for the small multirotor UAV, which could perform better than the standard RC joystick (RCJ) controller, and to evaluate the performance of the proposed interface. The multimodal speech and visual gesture (mSVG)interface were proposed, designed, and developed. This was then coupled to a Rotor S ROS Gazebo UAV simulator. An experiment study was designed to determine how practical the use of the proposed multimodal speech and visual gesture interface was in the control of small multirotor UAVs by determining the limits of speech and gesture at different ambient noise levels and under different background-lighting conditions, respectively. And to determine how the mSVG interface compares to the RC joystick controller for a simple navigational control task - in terms of performance (time of completion and accuracy of navigational control) and from a human factor?s perspective (user satisfaction and cognitive workload). 37 participants were recruited. From the results of the experiments conducted, the mSVG interface was found to be an effective alternative to the RCJ interface when operated within a constrained application environment. From the result of the noise level experiment, it was observed that speech recognition accuracy/success rate falls as noise levels rise, with75 dB noise level being the practical aerial robot (aerobot) application limit. From the results of the gesture lighting experiment, gestures were successfully recognised from 10 Lux and above on distinct solid backgrounds, but the effect of varying both the lighting conditions and the environment background on the quality of gesture recognition, was insignificant (< 0.5%), implying that the technology used, type of gesture captured, and the image processing technique used were more important. From the result of the performance and cognitive workload comparison between the RCJ and mSVG interfaces, the mSVG interface was found to perform better at higher nCA application levels than the RCJ interface. The mSVG interface was 1 minute faster and 25% more accurate than the RCJ interface; and the RCJ interface was found to be 1.4 times more cognitively demanding than the mSVG interface. The main limitation of this research was the limited lighting level range of 10 Lux - 1400 Lux used during the gesture lighting experiment, which constrains the application limit to lowlighting indoor environments. Suggested further works from this research included the development of a more robust gesture and speech algorithm and the coupling of the improved mSVG interface on to a practical UAV.ensuremath</pensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {phdthesis}
}
ensuremath<p class="MsoNormal"ensuremath>This research conducted an investigation into the use of novel human computer interaction(HCI) interfaces in the control of small multirotor unmanned aerial vehicles(UAVs). The main objective was to propose, design, and develop an alternative control interface for the small multirotor UAV, which could perform better than the standard RC joystick (RCJ) controller, and to evaluate the performance of the proposed interface. The multimodal speech and visual gesture (mSVG)interface were proposed, designed, and developed. This was then coupled to a Rotor S ROS Gazebo UAV simulator. An experiment study was designed to determine how practical the use of the proposed multimodal speech and visual gesture interface was in the control of small multirotor UAVs by determining the limits of speech and gesture at different ambient noise levels and under different background-lighting conditions, respectively. And to determine how the mSVG interface compares to the RC joystick controller for a simple navigational control task - in terms of performance (time of completion and accuracy of navigational control) and from a human factor?s perspective (user satisfaction and cognitive workload). 37 participants were recruited. From the results of the experiments conducted, the mSVG interface was found to be an effective alternative to the RCJ interface when operated within a constrained application environment. From the result of the noise level experiment, it was observed that speech recognition accuracy/success rate falls as noise levels rise, with75 dB noise level being the practical aerial robot (aerobot) application limit. From the results of the gesture lighting experiment, gestures were successfully recognised from 10 Lux and above on distinct solid backgrounds, but the effect of varying both the lighting conditions and the environment background on the quality of gesture recognition, was insignificant (< 0.5%), implying that the technology used, type of gesture captured, and the image processing technique used were more important. From the result of the performance and cognitive workload comparison between the RCJ and mSVG interfaces, the mSVG interface was found to perform better at higher nCA application levels than the RCJ interface. The mSVG interface was 1 minute faster and 25% more accurate than the RCJ interface; and the RCJ interface was found to be 1.4 times more cognitively demanding than the mSVG interface. The main limitation of this research was the limited lighting level range of 10 Lux - 1400 Lux used during the gesture lighting experiment, which constrains the application limit to lowlighting indoor environments. Suggested further works from this research included the development of a more robust gesture and speech algorithm and the coupling of the improved mSVG interface on to a practical UAV.ensuremath</pensuremath>
Abioye, Ayodeji; Naiseh, Mohammad; Hunt, William; Clark, Jediah R; Ramchurn, Sarvapali D; Soorati, Mohammad
The effect of data visualisation quality and task density on human-swarm interaction Proceedings Article
In: Proceedings of the 2023 32nd IEEE International Conference on Robot and Human Interactive Communication (RO-MAN), IEEE, 2023.
Abstract | Links | BibTeX | Tags: featured_publication
@inproceedings{soton479970,
title = {The effect of data visualisation quality and task density on human-swarm interaction},
author = {Ayodeji Abioye and Mohammad Naiseh and William Hunt and Jediah R Clark and Sarvapali D Ramchurn and Mohammad Soorati},
url = {https://eprints.soton.ac.uk/479970/},
year = {2023},
date = {2023-06-01},
urldate = {2023-06-01},
booktitle = {Proceedings of the 2023 32nd IEEE International Conference on Robot and Human Interactive Communication (RO-MAN)},
publisher = {IEEE},
abstract = {Despite the advantages of having robot swarms, human supervision is required for real-world applications. The performance of the human-swarm system depends on several factors including the data availability for the human operators. In this paper, we study the human factors aspect of the human-swarm interaction and investigate how having access to high-quality data can affect the performance of the human-swarm system - the number of tasks completed and the human trust level in operation. We designed an experiment where a human operator is tasked to operate a swarm to identify casualties in an area within a given time period. One group of operators had the option to request high-quality pictures while the other group had to base their decision on the available low-quality images. We performed a user study with 120 participants and recorded their success rate (directly logged via the simulation platform) as well as their workload and trust level (measured through a questionnaire after completing a human-swarm scenario). The findings from our study indicated that the group granted access to high-quality data exhibited an increased workload and placed greater trust in the swarm, thus confirming our initial hypothesis. However, we also found that the number of accurately identified casualties did not significantly vary between the two groups, suggesting that data quality had no impact on the successful completion of tasks.},
keywords = {featured_publication},
pubstate = {published},
tppubtype = {inproceedings}
}
Despite the advantages of having robot swarms, human supervision is required for real-world applications. The performance of the human-swarm system depends on several factors including the data availability for the human operators. In this paper, we study the human factors aspect of the human-swarm interaction and investigate how having access to high-quality data can affect the performance of the human-swarm system - the number of tasks completed and the human trust level in operation. We designed an experiment where a human operator is tasked to operate a swarm to identify casualties in an area within a given time period. One group of operators had the option to request high-quality pictures while the other group had to base their decision on the available low-quality images. We performed a user study with 120 participants and recorded their success rate (directly logged via the simulation platform) as well as their workload and trust level (measured through a questionnaire after completing a human-swarm scenario). The findings from our study indicated that the group granted access to high-quality data exhibited an increased workload and placed greater trust in the swarm, thus confirming our initial hypothesis. However, we also found that the number of accurately identified casualties did not significantly vary between the two groups, suggesting that data quality had no impact on the successful completion of tasks.
Krook, Joshua; McAuley, Derek; Anderson, Stuart; Downer, John; Winter, Peter; Ramchurn, Sarvapali D
AI Foundation Models: initial review, CMA Consultation, TAS Hub Response Miscellaneous
2023.
Links | BibTeX | Tags: Artificial Intelligence, Competition policy, featured_publication, Foundation Models, Large Language Models, markets
@misc{soton477553,
title = {AI Foundation Models: initial review, CMA Consultation, TAS Hub Response},
author = {Joshua Krook and Derek McAuley and Stuart Anderson and John Downer and Peter Winter and Sarvapali D Ramchurn},
url = {https://eprints.soton.ac.uk/477553/},
year = {2023},
date = {2023-06-01},
urldate = {2023-06-01},
publisher = {University of Southampton},
keywords = {Artificial Intelligence, Competition policy, featured_publication, Foundation Models, Large Language Models, markets},
pubstate = {published},
tppubtype = {misc}
}
Krook, Joshua; Downer, John; Winter, Peter; Williams, Jennifer; Ives, Jonathan; Bratu, Roxana; Sheir, Stephanie; Williams, Robin; Anderson, Stuart; Li, Phoebe; Ramamoorthy, Subramanian; Ramchurn, Sarvapali
AI regulation: a pro-innovation approach ? policy proposals: TASHub Response Miscellaneous
2023.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Consultation, innovation, Regulation, Trustworthy Autonomous Systems
@misc{soton478329,
title = {AI regulation: a pro-innovation approach ? policy proposals: TASHub Response},
author = {Joshua Krook and John Downer and Peter Winter and Jennifer Williams and Jonathan Ives and Roxana Bratu and Stephanie Sheir and Robin Williams and Stuart Anderson and Phoebe Li and Subramanian Ramamoorthy and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/478329/},
year = {2023},
date = {2023-06-01},
publisher = {University of Southampton},
abstract = {Response to open consultation from: Department for Science, Innovation and Technologyensuremath<br/ensuremath>and Office for Artificial Intelligence},
keywords = {Artificial Intelligence, Consultation, innovation, Regulation, Trustworthy Autonomous Systems},
pubstate = {published},
tppubtype = {misc}
}
Response to open consultation from: Department for Science, Innovation and Technologyensuremath<br/ensuremath>and Office for Artificial Intelligence
Hunt, William; Ryan, Jack; Abioye, Ayodeji O; Ramchurn, Sarvapali D; Soorati, Mohammad D
Demonstrating performance benefits of human-swarm teaming Proceedings Article
In: Proceedings of the 2023 International Conference on Autonomous Agents and Multiagent Systems, pp. 3062–3064, International Foundation for Autonomous Agents and Multiagent Systems (IFAAMAS), 2023.
Abstract | Links | BibTeX | Tags: Applications
@inproceedings{soton479903,
title = {Demonstrating performance benefits of human-swarm teaming},
author = {William Hunt and Jack Ryan and Ayodeji O Abioye and Sarvapali D Ramchurn and Mohammad D Soorati},
url = {https://eprints.soton.ac.uk/479903/},
year = {2023},
date = {2023-05-01},
urldate = {2023-05-01},
booktitle = {Proceedings of the 2023 International Conference on Autonomous Agents and Multiagent Systems},
pages = {3062–3064},
publisher = {International Foundation for Autonomous Agents and Multiagent Systems (IFAAMAS)},
abstract = {Autonomous swarms of robots can bring robustness, scalability and adaptability to safety-critical tasks such as search and rescue but their application is still very limited. Using semi-autonomous swarms with human control can bring robot swarms to real-world applications. Human operators can define goals for the swarm, monitor their performance and interfere with, or overrule, the decisions and behaviour. We present the "Human And Robot Interactive Swarm'' simulator (HARIS) that allows multi-user interaction with a robot swarm and facilitates qualitative and quantitative user studies through simulation of robot swarms completing tasks, from package delivery to search and rescue, with varying levels of human control. In this demonstration, we showcase the simulator by using it to study the performance gain offered by maintaining a "human-in-the-loop'' over a fully autonomous system as an example. This is illustrated in the context of search and rescue, with an autonomous allocation of resources to those in need.},
keywords = {Applications},
pubstate = {published},
tppubtype = {inproceedings}
}
Autonomous swarms of robots can bring robustness, scalability and adaptability to safety-critical tasks such as search and rescue but their application is still very limited. Using semi-autonomous swarms with human control can bring robot swarms to real-world applications. Human operators can define goals for the swarm, monitor their performance and interfere with, or overrule, the decisions and behaviour. We present the "Human And Robot Interactive Swarm'' simulator (HARIS) that allows multi-user interaction with a robot swarm and facilitates qualitative and quantitative user studies through simulation of robot swarms completing tasks, from package delivery to search and rescue, with varying levels of human control. In this demonstration, we showcase the simulator by using it to study the performance gain offered by maintaining a "human-in-the-loop'' over a fully autonomous system as an example. This is illustrated in the context of search and rescue, with an autonomous allocation of resources to those in need.
Worrawichaipat, Phuriwat; Gerding, Enrico; Kaparias, Ioannis; Ramchurn, Sarvapali
Multi-agent signal-less intersection management with dynamic platoon formation Proceedings Article
In: 22nd International Conference on Autonomous Agents and Multiagent Systems (29/05/23 - 02/06/23), pp. 1542–1550, 2023.
Links | BibTeX | Tags: featured_publication
@inproceedings{soton478647,
title = {Multi-agent signal-less intersection management with dynamic platoon formation},
author = {Phuriwat Worrawichaipat and Enrico Gerding and Ioannis Kaparias and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/478647/},
year = {2023},
date = {2023-05-01},
urldate = {2023-05-01},
booktitle = {22nd International Conference on Autonomous Agents and Multiagent Systems (29/05/23 - 02/06/23)},
pages = {1542–1550},
keywords = {featured_publication},
pubstate = {published},
tppubtype = {inproceedings}
}
Everett, Gregory; Beal, Ryan J; Matthews, Tim; Early, Joseph; Norman, Timothy J; Ramchurn, Sarvapali D
Inferring player location in sports matches: multi-agent spatial imputation from limited observations Miscellaneous
2023.
Abstract | Links | BibTeX | Tags: cs.LG, cs.MA
@misc{soton477020,
title = {Inferring player location in sports matches: multi-agent spatial imputation from limited observations},
author = {Gregory Everett and Ryan J Beal and Tim Matthews and Joseph Early and Timothy J Norman and Sarvapali D Ramchurn},
url = {https://eprints.soton.ac.uk/477020/},
year = {2023},
date = {2023-02-01},
abstract = {Understanding agent behaviour in Multi-Agent Systems (MAS) is an important problem in domains such as autonomous driving, disaster response, and sports analytics. Existing MAS problems typically use uniform timesteps with observations for all agents. In this work, we analyse the problem of agent location imputation, specifically posed in environments with non-uniform timesteps and limited agent observability (textttchar12695% missing values). Our approach uses Long Short-Term Memory and Graph Neural Network components to learn temporal and inter-agent patterns to predict the location of all agents at every timestep. We apply this to the domain of football (soccer) by imputing the location of all players in a game from sparse event data (e.g., shots and passes). Our model estimates player locations to within textttchar1266.9m; a textttchar12662% reduction in error from the best performing baseline. This approach facilitates downstream analysis tasks such as player physical metrics, player coverage, and team pitch control. Existing solutions to these tasks often require optical tracking data, which is expensive to obtain and only available to elite clubs. By imputing player locations from easy to obtain event data, we increase the accessibility of downstream tasks.},
keywords = {cs.LG, cs.MA},
pubstate = {published},
tppubtype = {misc}
}
Understanding agent behaviour in Multi-Agent Systems (MAS) is an important problem in domains such as autonomous driving, disaster response, and sports analytics. Existing MAS problems typically use uniform timesteps with observations for all agents. In this work, we analyse the problem of agent location imputation, specifically posed in environments with non-uniform timesteps and limited agent observability (textttchar12695% missing values). Our approach uses Long Short-Term Memory and Graph Neural Network components to learn temporal and inter-agent patterns to predict the location of all agents at every timestep. We apply this to the domain of football (soccer) by imputing the location of all players in a game from sparse event data (e.g., shots and passes). Our model estimates player locations to within textttchar1266.9m; a textttchar12662% reduction in error from the best performing baseline. This approach facilitates downstream analysis tasks such as player physical metrics, player coverage, and team pitch control. Existing solutions to these tasks often require optical tracking data, which is expensive to obtain and only available to elite clubs. By imputing player locations from easy to obtain event data, we increase the accessibility of downstream tasks.
2022
Ahmed, Sarah; Azim, Tayyaba; Early, Joseph Arthur; Ramchurn, Sarvapali
Revisiting deep fisher vectors: using fisher information to improve object classification Proceedings Article
In: British Machine Vision Conference (21/11/22 - 24/11/22), 2022.
Abstract | Links | BibTeX | Tags:
@inproceedings{soton471260,
title = {Revisiting deep fisher vectors: using fisher information to improve object classification},
author = {Sarah Ahmed and Tayyaba Azim and Joseph Arthur Early and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/471260/},
year = {2022},
date = {2022-11-01},
booktitle = {British Machine Vision Conference (21/11/22 - 24/11/22)},
abstract = {Although deep learning models have become the gold standard in achieving outstanding results on a large variety of computer vision and machine learning tasks, the use of kernel methods has still not gone out of trend because of its potential to beat deep learning performances at a number of occasions. Given the potential of kernel techniques, prior works have also proposed the use of hybrid approaches combining deep learning with kernel learning to complement their respective strengths and weaknesses. This work develops this idea further by introducing an improved version of Fisher kernels derived from the deep Boltzmann machines (DBM). Our improved deep Fisher kernel (IDFK) utilises an approximation of the Fisher information matrix to derive improved Fisher vectors. We show IDFK can be utilised to retain a high degree of class separability, making it appropriate for classification and retrieval tasks. The efficacy of the proposed approach is evaluated on three benchmark data sets: MNIST, USPS and Alphanumeric, showing an improvement in classification performance over existing kernel approaches, and comparable performance to deep learning methods, but with much reduced computational costs. Using explainable AI methods, we also demonstrate why our IDFK leads to better classification performance.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Although deep learning models have become the gold standard in achieving outstanding results on a large variety of computer vision and machine learning tasks, the use of kernel methods has still not gone out of trend because of its potential to beat deep learning performances at a number of occasions. Given the potential of kernel techniques, prior works have also proposed the use of hybrid approaches combining deep learning with kernel learning to complement their respective strengths and weaknesses. This work develops this idea further by introducing an improved version of Fisher kernels derived from the deep Boltzmann machines (DBM). Our improved deep Fisher kernel (IDFK) utilises an approximation of the Fisher information matrix to derive improved Fisher vectors. We show IDFK can be utilised to retain a high degree of class separability, making it appropriate for classification and retrieval tasks. The efficacy of the proposed approach is evaluated on three benchmark data sets: MNIST, USPS and Alphanumeric, showing an improvement in classification performance over existing kernel approaches, and comparable performance to deep learning methods, but with much reduced computational costs. Using explainable AI methods, we also demonstrate why our IDFK leads to better classification performance.
Yazdanpanah, Vahid; Gerding, Enrico; Stein, Sebastian; Dastani, Mehdi; Jonker, Catholijn M; Norman, Timothy; Ramchurn, Sarvapali
Reasoning About Responsibility in Autonomous Systems: Challenges and Opportunities Journal Article
In: AI & Society, 2022.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Citizen-Centric AI Systems, human-agent collectives, Human-Centred AI, Multiagent Responsibility Reasoning, Multiagent Systems, Trustworthy Autonomous Systems
@article{soton471971,
title = {Reasoning About Responsibility in Autonomous Systems: Challenges and Opportunities},
author = {Vahid Yazdanpanah and Enrico Gerding and Sebastian Stein and Mehdi Dastani and Catholijn M Jonker and Timothy Norman and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/471971/},
year = {2022},
date = {2022-11-01},
journal = {AI & Society},
abstract = {Ensuring the trustworthiness of autonomous systems and artificial intelligenceensuremath<br/ensuremath>is an important interdisciplinary endeavour. In this position paper, we argue thatensuremath<br/ensuremath>this endeavour will benefit from technical advancements in capturing various forms of responsibility, and we present a comprehensive research agenda to achieve this. In particular, we argue that ensuring the reliability of autonomous system can take advantage of technical approaches for quantifying degrees of responsibility and for coordinating tasks based on that. Moreover, we deem that, in certifying the legality of an AI system, formal and computationally implementable notions of responsibility, blame, accountability, and liability are applicable for addressing potential responsibility gaps (i.e., situations in which a group is responsible, but individuals? responsibility may be unclear). This is a call to enable AI systems themselves, as well as those involved in the design, monitoring, and governance of AI systems, to represent and reason about who can be seen as responsible in prospect (e.g., for completing a task in future) and who can be seen as responsible retrospectively (e.g., for a failure that has already occurred). To that end, in this work, we show that across all stages of the design, development, and deployment of Trustworthy Autonomous Systems (TAS), responsibility reasoning should play a key role. This position paper is the first step towards establishing a road-map and research agenda on how the notion of responsibility can provide novel solution concepts for ensuring the reliability and legality of TAS and, as a result, enables an effective embedding of AI technologies into society.},
keywords = {Artificial Intelligence, Citizen-Centric AI Systems, human-agent collectives, Human-Centred AI, Multiagent Responsibility Reasoning, Multiagent Systems, Trustworthy Autonomous Systems},
pubstate = {published},
tppubtype = {article}
}
Ensuring the trustworthiness of autonomous systems and artificial intelligenceensuremath<br/ensuremath>is an important interdisciplinary endeavour. In this position paper, we argue thatensuremath<br/ensuremath>this endeavour will benefit from technical advancements in capturing various forms of responsibility, and we present a comprehensive research agenda to achieve this. In particular, we argue that ensuring the reliability of autonomous system can take advantage of technical approaches for quantifying degrees of responsibility and for coordinating tasks based on that. Moreover, we deem that, in certifying the legality of an AI system, formal and computationally implementable notions of responsibility, blame, accountability, and liability are applicable for addressing potential responsibility gaps (i.e., situations in which a group is responsible, but individuals? responsibility may be unclear). This is a call to enable AI systems themselves, as well as those involved in the design, monitoring, and governance of AI systems, to represent and reason about who can be seen as responsible in prospect (e.g., for completing a task in future) and who can be seen as responsible retrospectively (e.g., for a failure that has already occurred). To that end, in this work, we show that across all stages of the design, development, and deployment of Trustworthy Autonomous Systems (TAS), responsibility reasoning should play a key role. This position paper is the first step towards establishing a road-map and research agenda on how the notion of responsibility can provide novel solution concepts for ensuring the reliability and legality of TAS and, as a result, enables an effective embedding of AI technologies into society.
Early, Joseph; Deweese, Ying-Jung; Evers, Christine; Ramchurn, Sarvapali
Scene-to-Patch earth observation: multiple instance learning for land cover classification Miscellaneous
2022, (14 pages total (4 main content; 2 acknowledgments + citations; 8 appendices); 8 figures (2 main; 6 appendix); published at "Tackling Climate Change with Machine Learning: Workshop at NeurIPS 2022").
Abstract | Links | BibTeX | Tags: cs.CV, cs.LG
@misc{soton472853,
title = {Scene-to-Patch earth observation: multiple instance learning for land cover classification},
author = {Joseph Early and Ying-Jung Deweese and Christine Evers and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/472853/},
year = {2022},
date = {2022-11-01},
abstract = {Land cover classification (LCC), and monitoring how land use changes over time, is an important process in climate change mitigation and adaptation. Existing approaches that use machine learning with Earth observation data for LCC rely on fully-annotated and segmented datasets. Creating these datasets requires a large amount of effort, and a lack of suitable datasets has become an obstacle in scaling the use of LCC. In this study, we propose Scene-to-Patch models: an alternative LCC approach utilising Multiple Instance Learning (MIL) that requires only high-level scene labels. This enables much faster development of new datasets whilst still providing segmentation through patch-level predictions, ultimately increasing the accessibility of using LCC for different scenarios. On the DeepGlobe-LCC dataset, our approach outperforms non-MIL baselines on both scene- and patch-level prediction. This work provides the foundation for expanding the use of LCC in climate change mitigation methods for technology, government, and academia.},
note = {14 pages total (4 main content; 2 acknowledgments + citations; 8 appendices); 8 figures (2 main; 6 appendix); published at "Tackling Climate Change with Machine Learning: Workshop at NeurIPS 2022"},
keywords = {cs.CV, cs.LG},
pubstate = {published},
tppubtype = {misc}
}
Land cover classification (LCC), and monitoring how land use changes over time, is an important process in climate change mitigation and adaptation. Existing approaches that use machine learning with Earth observation data for LCC rely on fully-annotated and segmented datasets. Creating these datasets requires a large amount of effort, and a lack of suitable datasets has become an obstacle in scaling the use of LCC. In this study, we propose Scene-to-Patch models: an alternative LCC approach utilising Multiple Instance Learning (MIL) that requires only high-level scene labels. This enables much faster development of new datasets whilst still providing segmentation through patch-level predictions, ultimately increasing the accessibility of using LCC for different scenarios. On the DeepGlobe-LCC dataset, our approach outperforms non-MIL baselines on both scene- and patch-level prediction. This work provides the foundation for expanding the use of LCC in climate change mitigation methods for technology, government, and academia.
Parnell, Katie; Fischer, Joel E; Clark, Jediah R; Bodenmann, Adrian; Trigo, Maria Jose Galvez; Brito, Mario; Soorati, Mohammad Divband; Plant, Katherine; Ramchurn, Sarvapali
Trustworthy UAV relationships: Applying the Schema Action World taxonomy to UAVs and UAV swarm operations Journal Article
In: International Journal of Human-Computer Interaction, 2022.
Abstract | Links | BibTeX | Tags:
@article{soton468839,
title = {Trustworthy UAV relationships: Applying the Schema Action World taxonomy to UAVs and UAV swarm operations},
author = {Katie Parnell and Joel E Fischer and Jediah R Clark and Adrian Bodenmann and Maria Jose Galvez Trigo and Mario Brito and Mohammad Divband Soorati and Katherine Plant and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/468839/},
year = {2022},
date = {2022-07-01},
journal = {International Journal of Human-Computer Interaction},
abstract = {Human Factors play a significant role inthe development and integration of avionic systems to ensure that they are trusted and can be used effectively. As Unoccupied Aerial Vehicle (UAV) technology becomes increasingly important to the aviation domain this holds true. This study aims to gain an understanding of UAV operators?trust requirements when piloting UAVs by utilising a popular aviation interview methodology (Schema World Action Research Method), in combination with key questions on trust identified from the literature. Interviews were conducted with six UAVoperators, with a range of experience. This identified the importance of past experience to trust and the expectations that operators hold. Recommendations are made that target training to inform experience, in addition to the equipment, procedures and organisational standards that can aid in developing trustworthy systems. The methodology that was developed shows promise for capturing trust within human-automation interactions},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Human Factors play a significant role inthe development and integration of avionic systems to ensure that they are trusted and can be used effectively. As Unoccupied Aerial Vehicle (UAV) technology becomes increasingly important to the aviation domain this holds true. This study aims to gain an understanding of UAV operators?trust requirements when piloting UAVs by utilising a popular aviation interview methodology (Schema World Action Research Method), in combination with key questions on trust identified from the literature. Interviews were conducted with six UAVoperators, with a range of experience. This identified the importance of past experience to trust and the expectations that operators hold. Recommendations are made that target training to inform experience, in addition to the equipment, procedures and organisational standards that can aid in developing trustworthy systems. The methodology that was developed shows promise for capturing trust within human-automation interactions
1.
Singh, Lokesh; Deshmukh, Jayati; Georgara, Athina; Nguyen, Tan Viet Tuyen; Ramchurn, Gopal
CareOps: A multi agent control room for independent living with care Proceedings Article
In: 25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26), pp. 4146–4148, 2026.
@inproceedings{soton511472,
title = {CareOps: A multi agent control room for independent living with care},
author = {Lokesh Singh and Jayati Deshmukh and Athina Georgara and Tan Viet Tuyen Nguyen and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/511472/},
year = {2026},
date = {2026-05-01},
booktitle = {25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26)},
pages = {4146–4148},
abstract = {This paper introduces CareOps, a multi-agent control-room dashboard for Independent Living with Care (ILWC) that coordinates heterogeneous sensors and a Buddy robot across multiple simulated homes. Agents fuse radar, audio, bed, passive infrared (PIR), and gait data into single prioritised incidents with human-readable explanations, and support one-click dispatch of either a robot (via socket) or a human carer (via smartphone notification). The demo presents a single decision-support workflow that orchestrates sensors, software agents, and robots through a unified interface. It provides a human-in-the-loop platform for exploring how professional care staff understand multi-sensor evidence and choose between robot and human responses.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper introduces CareOps, a multi-agent control-room dashboard for Independent Living with Care (ILWC) that coordinates heterogeneous sensors and a Buddy robot across multiple simulated homes. Agents fuse radar, audio, bed, passive infrared (PIR), and gait data into single prioritised incidents with human-readable explanations, and support one-click dispatch of either a robot (via socket) or a human carer (via smartphone notification). The demo presents a single decision-support workflow that orchestrates sensors, software agents, and robots through a unified interface. It provides a human-in-the-loop platform for exploring how professional care staff understand multi-sensor evidence and choose between robot and human responses.
2.
Georgara, Athina; Deshmukh, Jayati; Ramchurn, Gopal
GreenLine: A delay-tolerant mechanism design for grid capacity allocation Proceedings Article
In: 25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26), pp. 3323–3325, 2026.
@inproceedings{soton511473,
title = {GreenLine: A delay-tolerant mechanism design for grid capacity allocation},
author = {Athina Georgara and Jayati Deshmukh and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/511473/},
year = {2026},
date = {2026-05-01},
booktitle = {25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26)},
pages = {3323–3325},
abstract = {This work presents GreenLine, a new auction-based mechanism for integrating renewable power plants (RPP) in the power grid. We address a challenging task, where RPP owners want to maximise their profits by installing new power plants in the grid; while, the Distribution Network Operator (DNO) seeks to maximise the generated power while reducing potential installation delays due to growing power demand. This paper formulates this task as a multi-agent system, studies its useful properties such as incentive compatibility, individual rationality and economical efficiency, and discusses GreenLine under different deployment variations.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This work presents GreenLine, a new auction-based mechanism for integrating renewable power plants (RPP) in the power grid. We address a challenging task, where RPP owners want to maximise their profits by installing new power plants in the grid; while, the Distribution Network Operator (DNO) seeks to maximise the generated power while reducing potential installation delays due to growing power demand. This paper formulates this task as a multi-agent system, studies its useful properties such as incentive compatibility, individual rationality and economical efficiency, and discusses GreenLine under different deployment variations.
3.
Deshmukh, Jayati; Yazdanpanah, Vahid; Stein, Sebastian; Ramchurn, Gopal
The triad of identity, trust and responsibility in multi-agent systems Proceedings Article
In: 25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26), 2026.
@inproceedings{soton509930,
title = {The triad of identity, trust and responsibility in multi-agent systems},
author = {Jayati Deshmukh and Vahid Yazdanpanah and Sebastian Stein and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/509930/},
year = {2026},
date = {2026-05-01},
booktitle = {25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26)},
abstract = {The design of autonomous AI agents that behave responsibly and foster trust in open multi-agent systems remains a fundamental challenge. Traditional game-theoretical approaches largely assume self-interested behaviour, yet real-world collaborations among humans often rely on prosocial considerations that extend beyond individual utility. To address this, for the first time in this paper, we investigate the triad of identity, responsibility, and trust as core elements shaping responsible multi-agent behaviour. We propose a novel agent model, building on the notion of Computational Transcendence, which equips agents with an elastic sense of identity, enabling them to incorporate the welfare of others into their decision-making. Our framework integrates subjective (identity-based) and objective (experience-based and reputation-based) components of trust. Using Iterated Prisoner?s Dilemma (IPD) simulations on different network structures, we analyse how varying levels of identity and trust affect responsible behaviour. Results demonstrate that the interplay of these three concepts can promote emergent responsibility, mitigate exploitation, and sustain long-term cooperation in dynamic multi-agent environments. We argue that this triadic perspective provides a principled foundation for designing trustworthy, responsible, and identity/value aware agents with implications for future human?AI collaboration.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The design of autonomous AI agents that behave responsibly and foster trust in open multi-agent systems remains a fundamental challenge. Traditional game-theoretical approaches largely assume self-interested behaviour, yet real-world collaborations among humans often rely on prosocial considerations that extend beyond individual utility. To address this, for the first time in this paper, we investigate the triad of identity, responsibility, and trust as core elements shaping responsible multi-agent behaviour. We propose a novel agent model, building on the notion of Computational Transcendence, which equips agents with an elastic sense of identity, enabling them to incorporate the welfare of others into their decision-making. Our framework integrates subjective (identity-based) and objective (experience-based and reputation-based) components of trust. Using Iterated Prisoner?s Dilemma (IPD) simulations on different network structures, we analyse how varying levels of identity and trust affect responsible behaviour. Results demonstrate that the interplay of these three concepts can promote emergent responsibility, mitigate exploitation, and sustain long-term cooperation in dynamic multi-agent environments. We argue that this triadic perspective provides a principled foundation for designing trustworthy, responsible, and identity/value aware agents with implications for future human?AI collaboration.
4.
Everett, Gregory Alan
Models and algorithms to optimise team performance in football PhD Thesis
University of Southampton, 2026.
@phdthesis{soton509865,
title = {Models and algorithms to optimise team performance in football},
author = {Gregory Alan Everett},
url = {https://eprints.soton.ac.uk/509865/},
year = {2026},
date = {2026-03-01},
publisher = {University of Southampton},
school = {University of Southampton},
abstract = {With the growing use of Artificial Intelligence (AI) and machine learning, there is increasing potential to model and optimise team behaviour across important domains such as disaster response, security, and team sports. These domains are inherently spatiotemporal, requiring models that capture both spatial and temporal dynamics. This thesis focuses on football, a complex, dynamic sport with rich spatiotemporal data and clear objectives, making it an ideal testbed for developing and validating team-based AI models. Moreover, football analytics is a rapidly expanding industry, with European clubs generating ?38 billion in revenue during the 2023/24 season, driving the demand for models that provide competitive advantages.ensuremath<br/ensuremath>ensuremath<br/ensuremath>This thesis proposes a number of novel methods that utilise spatiotemporal data to advance team prediction, analysis, and decision-making in football and other team-based domains. In particular, we introduce a spatiotemporal agent behaviour imputation model that reduces predictive error by 62% compared to baselines in limited observability settings, significantly improving the accessibility of off-ball football analytics through imputed tracking data. We also present a novel spatial teamwork model, combining Monte Carlo tree search (MCTS) and linear programming, that optimises agent decision-making in real-world football defence, reducing opponent threat by 24%. In addition, we develop new metrics, derived from a graph attention network (GAT), to assign credit to indirect agent contributions in team-based defence. The GAT model predicts football passes with a 6% reduction in loss compared to baselines, and we show how these new metrics can greatly improve off-ball football player evaluation. Finally, we propose a dynamic team formation and agent replacement model that accounts for agent fatigue and unavailability and optimises decision-making using a multi-step MCTS algorithm. Applied to football team selection and substitutions, this model improves long-term team performance by 1% and reduces first-team injuries by 15%. This thesis also highlights key remaining challenges for AI in football, including the use of richer player data (e.g., body pose) and greater use of explainable models to build trust in clubs.},
keywords = {},
pubstate = {published},
tppubtype = {phdthesis}
}
With the growing use of Artificial Intelligence (AI) and machine learning, there is increasing potential to model and optimise team behaviour across important domains such as disaster response, security, and team sports. These domains are inherently spatiotemporal, requiring models that capture both spatial and temporal dynamics. This thesis focuses on football, a complex, dynamic sport with rich spatiotemporal data and clear objectives, making it an ideal testbed for developing and validating team-based AI models. Moreover, football analytics is a rapidly expanding industry, with European clubs generating ?38 billion in revenue during the 2023/24 season, driving the demand for models that provide competitive advantages.ensuremath<br/ensuremath>ensuremath<br/ensuremath>This thesis proposes a number of novel methods that utilise spatiotemporal data to advance team prediction, analysis, and decision-making in football and other team-based domains. In particular, we introduce a spatiotemporal agent behaviour imputation model that reduces predictive error by 62% compared to baselines in limited observability settings, significantly improving the accessibility of off-ball football analytics through imputed tracking data. We also present a novel spatial teamwork model, combining Monte Carlo tree search (MCTS) and linear programming, that optimises agent decision-making in real-world football defence, reducing opponent threat by 24%. In addition, we develop new metrics, derived from a graph attention network (GAT), to assign credit to indirect agent contributions in team-based defence. The GAT model predicts football passes with a 6% reduction in loss compared to baselines, and we show how these new metrics can greatly improve off-ball football player evaluation. Finally, we propose a dynamic team formation and agent replacement model that accounts for agent fatigue and unavailability and optimises decision-making using a multi-step MCTS algorithm. Applied to football team selection and substitutions, this model improves long-term team performance by 1% and reduces first-team injuries by 15%. This thesis also highlights key remaining challenges for AI in football, including the use of richer player data (e.g., body pose) and greater use of explainable models to build trust in clubs.
5.
Tuyen, Nguyen Tan Viet; Georgara, Athina; Singh, Lokesh; Deshmukh, Jayati; Davey, Sean; Tisdale, Paul N.; Ramchurn, Sarvapali
Supporting ?Stop and Watch? in elderly care with socially assistive robots: insights from a participatory design workshop Proceedings Article
In: Baillie, Lynne; Smart, William D.; Graaf, Maartje De; Gombolay, Matthew; Torre, Ilaria (Ed.): Companion Proceedings of the 21st ACM/IEEE International Conference on Human-Robot Interaction, HRI Companion 2026, pp. 302–306, ACM Press, 2026.
@inproceedings{soton511593,
title = {Supporting ?Stop and Watch? in elderly care with socially assistive robots: insights from a participatory design workshop},
author = {Nguyen Tan Viet Tuyen and Athina Georgara and Lokesh Singh and Jayati Deshmukh and Sean Davey and Paul N. Tisdale and Sarvapali Ramchurn},
editor = {Lynne Baillie and William D. Smart and Maartje De Graaf and Matthew Gombolay and Ilaria Torre},
url = {https://eprints.soton.ac.uk/511593/},
year = {2026},
date = {2026-03-01},
urldate = {2026-03-01},
booktitle = {Companion Proceedings of the 21st ACM/IEEE International Conference on Human-Robot Interaction, HRI Companion 2026},
pages = {302–306},
publisher = {ACM Press},
abstract = {ensuremath<pensuremath>?Stop and Watch? is an early-warning tool adopted by the UK NHS and is widely used in elderly care settings. The tool helps caregivers to recognise abnormal changes in residents? health. Despite its clinical value, the process remains highly manual, workload-intensive, and vulnerable to missed observations, particularly in environments facing staff shortages, frequent staff rotations, and increasing care demands. We argue that AI-based systems such as Socially Assistive Robots (SARs) and Assistive Living Technologies (ALTs) offer promising avenues for supporting and enhancing the ?Stop and Watch? tool. However, designing such systems requires a multidisciplinary effort to establish a comprehensive understanding of current practices and the priorities and concerns of all relevant stakeholders. This paper presents insights from a participatory design workshop held in a care home in the UK to explore how SARs and ALTs could meaningfully support the ?Stop and Watch? tool, understand stakeholders? expectations, perceived benefits, and concerns regarding deployment in this sensitive context.ensuremath</pensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
ensuremath<pensuremath>?Stop and Watch? is an early-warning tool adopted by the UK NHS and is widely used in elderly care settings. The tool helps caregivers to recognise abnormal changes in residents? health. Despite its clinical value, the process remains highly manual, workload-intensive, and vulnerable to missed observations, particularly in environments facing staff shortages, frequent staff rotations, and increasing care demands. We argue that AI-based systems such as Socially Assistive Robots (SARs) and Assistive Living Technologies (ALTs) offer promising avenues for supporting and enhancing the ?Stop and Watch? tool. However, designing such systems requires a multidisciplinary effort to establish a comprehensive understanding of current practices and the priorities and concerns of all relevant stakeholders. This paper presents insights from a participatory design workshop held in a care home in the UK to explore how SARs and ALTs could meaningfully support the ?Stop and Watch? tool, understand stakeholders? expectations, perceived benefits, and concerns regarding deployment in this sensitive context.ensuremath</pensuremath>
6.
Hafez, Muhammad Burhan; Miller, Emily; Milford, Michael J.; Ramchurn, Gopal; Ehsan, Shoaib
Through the lens of doubt: robust and efficient uncertainty estimation for visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 11, no. 5, pp. 5899–5906, 2026.
@article{soton510726,
title = {Through the lens of doubt: robust and efficient uncertainty estimation for visual place recognition},
author = {Muhammad Burhan Hafez and Emily Miller and Michael J. Milford and Gopal Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/510726/},
year = {2026},
date = {2026-03-01},
journal = {IEEE Robotics and Automation Letters},
volume = {11},
number = {5},
pages = {5899–5906},
abstract = {ensuremath<pensuremath>Visual Place Recognition (VPR) enables robots and autonomous vehicles to identify previously visited locations by matching current observations against a database of known places. However, VPR systems face significant challenges when deployed across varying visual environments, lighting conditions, seasonal changes, and viewpoints changes. Failure-critical VPR applications, such as loop closure detection in simultaneous localization and mapping (SLAM) pipelines, require robust estimation of place matching uncertainty. We propose three training-free uncertainty metrics that estimate prediction confidence by analyzing inherent statistical patterns in similarity scores from any existing VPR method. Similarity Distribution (SD) quantifies match distinctiveness by measuring score separation between candidates; Ratio Spread (RS) evaluates competitive ambiguity among top-scoring locations; and Statistical Uncertainty (SU) is a combination of SD and RS that provides a unified metric that generalizes across datasets and VPR methods without requiring validation data to select the optimal metric. All three metrics operate without additional model training, architectural modifications, or computationally expensive geometric verification. Comprehensive evaluation across nine state-of-the-art VPR methods and six benchmark datasets confirms that our metrics excel at discriminating between correct and incorrect VPR matches, and consistently outperform existing approaches while maintaining negligible computational overhead, making it deployable for real-time robotic applications across varied environmental conditions with improved precision-recall performance.ensuremath</pensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Visual Place Recognition (VPR) enables robots and autonomous vehicles to identify previously visited locations by matching current observations against a database of known places. However, VPR systems face significant challenges when deployed across varying visual environments, lighting conditions, seasonal changes, and viewpoints changes. Failure-critical VPR applications, such as loop closure detection in simultaneous localization and mapping (SLAM) pipelines, require robust estimation of place matching uncertainty. We propose three training-free uncertainty metrics that estimate prediction confidence by analyzing inherent statistical patterns in similarity scores from any existing VPR method. Similarity Distribution (SD) quantifies match distinctiveness by measuring score separation between candidates; Ratio Spread (RS) evaluates competitive ambiguity among top-scoring locations; and Statistical Uncertainty (SU) is a combination of SD and RS that provides a unified metric that generalizes across datasets and VPR methods without requiring validation data to select the optimal metric. All three metrics operate without additional model training, architectural modifications, or computationally expensive geometric verification. Comprehensive evaluation across nine state-of-the-art VPR methods and six benchmark datasets confirms that our metrics excel at discriminating between correct and incorrect VPR matches, and consistently outperform existing approaches while maintaining negligible computational overhead, making it deployable for real-time robotic applications across varied environmental conditions with improved precision-recall performance.ensuremath</pensuremath>
7.
Ramchurn, Gopal; Neff, Gina; Parisio, Isabela; Kiden, Sarah; Georgara, Athina; Stumpf, Simone; Wong, Mark; Shahandashti, Siamak F.; Aristodemou, Marios
Responsible AI UK response to the DSIT?s call for evidence on the AI Growth Lab Technical Report
no. 10.5258/RAi/005, 2026.
@techreport{soton508020,
title = {Responsible AI UK response to the DSIT?s call for evidence on the AI Growth Lab},
author = {Gopal Ramchurn and Gina Neff and Isabela Parisio and Sarah Kiden and Athina Georgara and Simone Stumpf and Mark Wong and Siamak F. Shahandashti and Marios Aristodemou},
url = {https://eprints.soton.ac.uk/508020/},
year = {2026},
date = {2026-01-01},
number = {10.5258/RAi/005},
publisher = {University of Southampton},
abstract = {A response to the Department for Science, Innovation and Technology (DSIT) open call for evidence regarding the AI Growth Lab1 on behalf of Responsible AI UK (RAi UK), an open and multidisciplinary network that brings together experts from across the four nations of the UK to understand how we should shape the development of AI to benefit people, communities and society},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
A response to the Department for Science, Innovation and Technology (DSIT) open call for evidence regarding the AI Growth Lab1 on behalf of Responsible AI UK (RAi UK), an open and multidisciplinary network that brings together experts from across the four nations of the UK to understand how we should shape the development of AI to benefit people, communities and society
8.
Owen, Aled Lloyd; Ramchurn, Gopal; Dasgupta, Prokar; Ao, Shuang; Barnard, Pepita; Deshmukh, Jayati; Parisio, Isabela; Singh, Lokesh; Valoor, Adarsh; Waheed, Maria; Braunschweig, Bertrand; Quintero, Karla; Poretschkin, Maximilian; Hawes, Ben; Khomh, Foutse
Advancing Trustworthy Artificial Intelligence: Lessons Learned and Emerging Challenges Proceedings Article
In: Responsible Ai, University of Southampton, 2025.
@inproceedings{soton506660,
title = {Advancing Trustworthy Artificial Intelligence: Lessons Learned and Emerging Challenges},
author = {Aled Lloyd Owen and Gopal Ramchurn and Prokar Dasgupta and Shuang Ao and Pepita Barnard and Jayati Deshmukh and Isabela Parisio and Lokesh Singh and Adarsh Valoor and Maria Waheed and Bertrand Braunschweig and Karla Quintero and Maximilian Poretschkin and Ben Hawes and Foutse Khomh},
url = {https://eprints.soton.ac.uk/506660/},
year = {2025},
date = {2025-11-01},
booktitle = {Responsible Ai},
publisher = {University of Southampton},
abstract = {Government white paper},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Government white paper
9.
Reyes-Cruz, Gisela; Kiden, Sarah; Azim, Tayyaba; Bergin, Aislinn Gomez; Choi, Sena; Eke, Damian; Klyshbekova, Maira; Peter, Oriane; Waheed, Maria; Devlin, Kate; Fischer, Joel; Vallejos, Elvira Perez; Ramchurn, Gopal; Stein, Sebastian
Response to call for input for EMRTD study ?Artificial Intelligence, Cultural Rights, and the Right to Development? Technical Report
no. 10.25878/7nmj-0t79, 2025.
@techreport{soton511244,
title = {Response to call for input for EMRTD study ?Artificial Intelligence, Cultural Rights, and the Right to Development?},
author = {Gisela Reyes-Cruz and Sarah Kiden and Tayyaba Azim and Aislinn Gomez Bergin and Sena Choi and Damian Eke and Maira Klyshbekova and Oriane Peter and Maria Waheed and Kate Devlin and Joel Fischer and Elvira Perez Vallejos and Gopal Ramchurn and Sebastian Stein},
url = {https://eprints.soton.ac.uk/511244/},
year = {2025},
date = {2025-11-01},
number = {10.25878/7nmj-0t79},
publisher = {University of Nottingham},
abstract = {Responsible AI UK's response to the Call for Input for EMRTD study "Artificial Intelligence, Cultural Rights, and the Right to Development" issued by the Expert Mechanism on the Right to Development United Nations Human Rights Office of the High Commissioner (OHCHR).ensuremath<br/ensuremath>ensuremath<br/ensuremath>https://www.ohchr.org/en/calls-for-input/2025/call-input-emrtd-study-artificial-intelligence-cultural-rights-and-right},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
Responsible AI UK's response to the Call for Input for EMRTD study "Artificial Intelligence, Cultural Rights, and the Right to Development" issued by the Expert Mechanism on the Right to Development United Nations Human Rights Office of the High Commissioner (OHCHR).ensuremath<br/ensuremath>ensuremath<br/ensuremath>https://www.ohchr.org/en/calls-for-input/2025/call-input-emrtd-study-artificial-intelligence-cultural-rights-and-right
10.
Ramchurn, Gopal; Owen, Aled Lloyd; Ao, Shuang; Barnard, Pepita; Bergin, Aislinn Gomez; Parisio, Isabela; Valoor, Adarsh; Waheed, Maria; Holter, Carolyn Ten; Portillo, Virginia; Procter, Rob; Batterham, Paul; Downer, John; Winter, Peter; Krook, Joshua; Blockx, Jan; Braunschweig, Bertrand; Quintero, Karla; Poretschkin, Maximilian; Becker, Daniel; Campbell-Ratcliffe, Emily; Hawes, Ben; Shaw, Patricia; Patel, Reema; Tewari, Ashish; Thomas, Alec; Duncan, Paul; Gillings, Eliot
Frameworks and Toolkits for Assuring Responsible AI Book Section
In: Responsible Ai UK, University of Southampton, 2025.
@incollection{soton506057,
title = {Frameworks and Toolkits for Assuring Responsible AI},
author = {Gopal Ramchurn and Aled Lloyd Owen and Shuang Ao and Pepita Barnard and Aislinn Gomez Bergin and Isabela Parisio and Adarsh Valoor and Maria Waheed and Carolyn Ten Holter and Virginia Portillo and Rob Procter and Paul Batterham and John Downer and Peter Winter and Joshua Krook and Jan Blockx and Bertrand Braunschweig and Karla Quintero and Maximilian Poretschkin and Daniel Becker and Emily Campbell-Ratcliffe and Ben Hawes and Patricia Shaw and Reema Patel and Ashish Tewari and Alec Thomas and Paul Duncan and Eliot Gillings},
url = {https://eprints.soton.ac.uk/506057/},
year = {2025},
date = {2025-10-01},
urldate = {2025-10-01},
booktitle = {Responsible Ai UK},
publisher = {University of Southampton},
abstract = {This report explores experiences of using tools for assuring responsible practice in the build, deployment, and governance of AI systems, in particular frameworks and toolkits developed and published by RAi UK-funded projects and by other organisations. We examine evidence of how and why these tools are used, and of their usefulness and their limitations. We look for lessons that could improve AI assurance in the future, including potentially using AI-based tools to support AI assurance. The aim is to identify priority areas and key questions for RAi UK and other researchers and organisations to explore further, with the goal of improving the fit between the supply of, and demand for, tools that support organisational assurance of responsible AI .Responsible Ai UK and our partner for this event, Confiance.ai, brought people together from across government, public services, and industry, as well as project teams that develop support for responsible AI. The aim was to clarify how people should approach the development and use of toolkits and/or frameworks in practice. We wanted to synthesise findings from our research across the programme and to find and address any gaps. The workshop took place online on Thursday 10 April 2025.},
keywords = {},
pubstate = {published},
tppubtype = {incollection}
}
This report explores experiences of using tools for assuring responsible practice in the build, deployment, and governance of AI systems, in particular frameworks and toolkits developed and published by RAi UK-funded projects and by other organisations. We examine evidence of how and why these tools are used, and of their usefulness and their limitations. We look for lessons that could improve AI assurance in the future, including potentially using AI-based tools to support AI assurance. The aim is to identify priority areas and key questions for RAi UK and other researchers and organisations to explore further, with the goal of improving the fit between the supply of, and demand for, tools that support organisational assurance of responsible AI .Responsible Ai UK and our partner for this event, Confiance.ai, brought people together from across government, public services, and industry, as well as project teams that develop support for responsible AI. The aim was to clarify how people should approach the development and use of toolkits and/or frameworks in practice. We wanted to synthesise findings from our research across the programme and to find and address any gaps. The workshop took place online on Thursday 10 April 2025.
11.
Kiden, Sarah; Peter, Oriane; Reyes-Cruz, Gisela; Klyshbekova, Maira; Choi, Sena; Bergin, Aislinn Gomez; Waheed, Maria; Eke, Damian; Azim, Tayyaba; Ramchurn, Sarvapali; Stein, Sebastian; Vallejos, Elvira Perez; Devlin, Kate; Fischer, Joel E.
Back to the communities: a mixed-methods and community-driven evaluation of cultural sensitivity in text-to-image models Miscellaneous
2025.
@misc{soton507303,
title = {Back to the communities: a mixed-methods and community-driven evaluation of cultural sensitivity in text-to-image models},
author = {Sarah Kiden and Oriane Peter and Gisela Reyes-Cruz and Maira Klyshbekova and Sena Choi and Aislinn Gomez Bergin and Maria Waheed and Damian Eke and Tayyaba Azim and Sarvapali Ramchurn and Sebastian Stein and Elvira Perez Vallejos and Kate Devlin and Joel E. Fischer},
url = {https://eprints.soton.ac.uk/507303/},
year = {2025},
date = {2025-10-01},
publisher = {arXiv},
abstract = {Evidence shows that text-to-image (T2I) models disproportionately reflect Western cultural norms, amplifying misrepresentation and harms to minority groups. However, evaluating cultural sensitivity is inherently complex due to its fluid and multifaceted nature. This paper draws on a state-of-the-art review and co-creation workshops involving 59 individuals from 19 different countries. We developed and validated a mixed-methods community-based evaluation methodology to assess cultural sensitivity in T2I models, which embraces first-person methods. Quantitative scores and qualitative inquiries expose convergence and disagreement within and across communities, illuminate the downstream consequences of misrepresentation, and trace how training data shaped by unequal power relations distort depictions. Extensive assessments are constrained by high resource requirements and the dynamic nature of culture, a tension we alleviate through a context-based and iterative methodology. The paper provides actionable recommendations for stakeholders, highlighting pathways to investigate the sources, mechanisms, and impacts of cultural (mis)representation in T2I models.},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Evidence shows that text-to-image (T2I) models disproportionately reflect Western cultural norms, amplifying misrepresentation and harms to minority groups. However, evaluating cultural sensitivity is inherently complex due to its fluid and multifaceted nature. This paper draws on a state-of-the-art review and co-creation workshops involving 59 individuals from 19 different countries. We developed and validated a mixed-methods community-based evaluation methodology to assess cultural sensitivity in T2I models, which embraces first-person methods. Quantitative scores and qualitative inquiries expose convergence and disagreement within and across communities, illuminate the downstream consequences of misrepresentation, and trace how training data shaped by unequal power relations distort depictions. Extensive assessments are constrained by high resource requirements and the dynamic nature of culture, a tension we alleviate through a context-based and iterative methodology. The paper provides actionable recommendations for stakeholders, highlighting pathways to investigate the sources, mechanisms, and impacts of cultural (mis)representation in T2I models.
12.
Ansari, Aamir Ahmad; Ramchurn, Gopal; Nguyen, Tan Viet Tuyen
Beyond text: multi-modal LLM in human robot interaction Proceedings Article
In: UK AI Research Symposium: A Festival of Ideas (08/09/25 - 09/09/25), 2025.
@inproceedings{soton505659,
title = {Beyond text: multi-modal LLM in human robot interaction},
author = {Aamir Ahmad Ansari and Gopal Ramchurn and Tan Viet Tuyen Nguyen},
url = {https://eprints.soton.ac.uk/505659/},
year = {2025},
date = {2025-09-01},
booktitle = {UK AI Research Symposium: A Festival of Ideas (08/09/25 - 09/09/25)},
abstract = {Multimodal interaction plays a vital role in Human-Robot Interaction (HRI), enabling robots to communicate with humans through multiple channels. This study introduces a novel approach to enhance such interactions by treating images and human motion as distinct foreign languages, in addition to text. In the proposed framework, vector quantization is employed to convert multimodal inputs such as images and human motions to an aligned set of tokens. A Large Language Model (LLM) is then pre-trained with the use of Low-Rank Adaptation (LoRA) and instruction-tuned on a dialogue dataset that incorporates both image and motion context. The proposed multimodal LLM framework aims to equip robots with the ability to understand and respond to complex human queries through multimodal inputs and outputs, enabling more natural and effective interactions.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Multimodal interaction plays a vital role in Human-Robot Interaction (HRI), enabling robots to communicate with humans through multiple channels. This study introduces a novel approach to enhance such interactions by treating images and human motion as distinct foreign languages, in addition to text. In the proposed framework, vector quantization is employed to convert multimodal inputs such as images and human motions to an aligned set of tokens. A Large Language Model (LLM) is then pre-trained with the use of Low-Rank Adaptation (LoRA) and instruction-tuned on a dialogue dataset that incorporates both image and motion context. The proposed multimodal LLM framework aims to equip robots with the ability to understand and respond to complex human queries through multimodal inputs and outputs, enabling more natural and effective interactions.
13.
Thavanesan, Navamayooran; Naiseh, Mohammad; Terol, Miguel; Rahman, Saqib Andrew; Hill, Samuel Luke; Parfitt, Charlotte; Walters, Zoë S; Ramchurn, Sarvapali; Markar, Sheraz; Owen, Richard; Maynard, Nick; Azim, Tayyaba; Belkatir, Zehor; Perez, Elvira Vallejos; McCord, Mimi; Underwood, Tim; Vigneswaran, Ganesh
The oesophageal cancer multi-disciplinary tool: a co-designed, externally validated, machine learning clinical decision support system Journal Article
In: EClinicalMedicine, vol. 89, pp. 103527, 2025, (For the purposes of open access, the authors have applied a Creative Commons attribution license (CC-BY) to any Author Accepted Manuscript version arising from this submission.).
@article{soton506614,
title = {The oesophageal cancer multi-disciplinary tool: a co-designed, externally validated, machine learning clinical decision support system},
author = {Navamayooran Thavanesan and Mohammad Naiseh and Miguel Terol and Saqib Andrew Rahman and Samuel Luke Hill and Charlotte Parfitt and Zoë S Walters and Sarvapali Ramchurn and Sheraz Markar and Richard Owen and Nick Maynard and Tayyaba Azim and Zehor Belkatir and Elvira Vallejos Perez and Mimi McCord and Tim Underwood and Ganesh Vigneswaran},
url = {https://eprints.soton.ac.uk/506614/},
year = {2025},
date = {2025-09-01},
journal = {EClinicalMedicine},
volume = {89},
pages = {103527},
abstract = {ensuremath<pensuremath>BACKGROUND: The oesophageal cancer (OC) multi-disciplinary team (MDT) operates under significant pressures, handling complex decision-making. Machine learning (ML) can learn complex decision-making paradigms to improve efficiency, consistency, and cost if trained and deployed responsibly. We present an externally validated ML-based clinical decision support system (CDSS) designed to predict OC MDT treatment decisions and prognosticate palliative scenarios, co-designed using Responsible Research and Innovation (RRI) principles.ensuremath</pensuremath>ensuremath<pensuremath>METHODS: Clinicopathological data collected from 1931 patients between 4th September 2009, and 8th November 2022 were used to test and validate models trained through four ML algorithms to predict curative and palliative treatment pathways along with palliative prognosis. 953 OC cases treated at University Hospitals Southampton (UHS) were used to train ML models which were externally validated on 978 OC cases from Oxford University Hospitals (OUH). Model performance was evaluated using Area Under Curve (AUC) for treatment classifiers and calibration curves for survival models. A parallel RRI program at the University of Southampton (United Kingdom) combining clinician interviews and inter-disciplinary workshops was conducted between 16.3.23 and 23.5.24. The RRI program comprised a group of 17 domain experts comprising programmers, computer scientists, clinicians and patient representatives to allow end-users to contribute towards the co-design of the CDSS user interface.ensuremath</pensuremath>ensuremath<pensuremath>FINDINGS: Cohorts differed in baseline characteristics, with the external cohort (OUH) being younger, having better performance status, and a higher prevalence of pulmonary and vascular disease. Despite these differences, on internal validation (UHS cohort) mean AUCs for the primary treatment model were: MLR 0.905 $±$ 0.048, XGB 0.909 $±$ 0.044 and RF 0.883 $±$ 0.059 (k = 5 cross-validation) and MLR 0.866 (95% CI 0.866-0.867), XGB 0.863 (0.862-0.864), RF 0.863 (0.867-0.868) on bootstrapped resampling. For the palliative classifier, mean AUCs were: MLR 0.805 $±$ 0.096, XGB 0.815 $±$ 0.081 and RF 0.793 $±$ 0.083 (k = 5 cross-validation) and MLR 0.736 (95% CI 0.734-0.737), XGB 0.799 (0.798-0.800), RF 0.781 (0.778-0.782) on bootstrapped resampling. On external validation (OUH cohort), AUCs were MLR 0.894, XGB 0.887 and RF 0.891 for the primary treatment model and MLR 0.711, XGB 0.742 and RF 0.730 for the palliative treatment classifier. Predicted survival probability from the palliative survival model was well calibrated over the first 12 months post-diagnosis in both cohorts. The RRI program provided a collaborative environment leading to valuable modifications to the CDSS including prediction explanations, visual aids for survival and integrated education for users producing a user-friendly and quick to use tool.ensuremath</pensuremath>ensuremath<pensuremath>INTERPRETATION: We present a novel, responsibly developed, externally validated AI CDSS trained to predict oesophageal cancer MDT decisions. It represents the foundations of a transformative application of ML, personalised, consistent and efficient MDT decision-support within OC which aligns to RRI principles.ensuremath</pensuremath>ensuremath<pensuremath>FUNDING: Doctoral Studentship for NT (Institute for Life Sciences (University of Southampton) & University Hospital Southampton), UKRI TAS Pump-Priming Grant (TAS_PP_00167).ensuremath</pensuremath>},
note = {For the purposes of open access, the authors have applied a Creative Commons attribution license (CC-BY) to any Author Accepted Manuscript version arising from this submission.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>BACKGROUND: The oesophageal cancer (OC) multi-disciplinary team (MDT) operates under significant pressures, handling complex decision-making. Machine learning (ML) can learn complex decision-making paradigms to improve efficiency, consistency, and cost if trained and deployed responsibly. We present an externally validated ML-based clinical decision support system (CDSS) designed to predict OC MDT treatment decisions and prognosticate palliative scenarios, co-designed using Responsible Research and Innovation (RRI) principles.ensuremath</pensuremath>ensuremath<pensuremath>METHODS: Clinicopathological data collected from 1931 patients between 4th September 2009, and 8th November 2022 were used to test and validate models trained through four ML algorithms to predict curative and palliative treatment pathways along with palliative prognosis. 953 OC cases treated at University Hospitals Southampton (UHS) were used to train ML models which were externally validated on 978 OC cases from Oxford University Hospitals (OUH). Model performance was evaluated using Area Under Curve (AUC) for treatment classifiers and calibration curves for survival models. A parallel RRI program at the University of Southampton (United Kingdom) combining clinician interviews and inter-disciplinary workshops was conducted between 16.3.23 and 23.5.24. The RRI program comprised a group of 17 domain experts comprising programmers, computer scientists, clinicians and patient representatives to allow end-users to contribute towards the co-design of the CDSS user interface.ensuremath</pensuremath>ensuremath<pensuremath>FINDINGS: Cohorts differed in baseline characteristics, with the external cohort (OUH) being younger, having better performance status, and a higher prevalence of pulmonary and vascular disease. Despite these differences, on internal validation (UHS cohort) mean AUCs for the primary treatment model were: MLR 0.905 $±$ 0.048, XGB 0.909 $±$ 0.044 and RF 0.883 $±$ 0.059 (k = 5 cross-validation) and MLR 0.866 (95% CI 0.866-0.867), XGB 0.863 (0.862-0.864), RF 0.863 (0.867-0.868) on bootstrapped resampling. For the palliative classifier, mean AUCs were: MLR 0.805 $±$ 0.096, XGB 0.815 $±$ 0.081 and RF 0.793 $±$ 0.083 (k = 5 cross-validation) and MLR 0.736 (95% CI 0.734-0.737), XGB 0.799 (0.798-0.800), RF 0.781 (0.778-0.782) on bootstrapped resampling. On external validation (OUH cohort), AUCs were MLR 0.894, XGB 0.887 and RF 0.891 for the primary treatment model and MLR 0.711, XGB 0.742 and RF 0.730 for the palliative treatment classifier. Predicted survival probability from the palliative survival model was well calibrated over the first 12 months post-diagnosis in both cohorts. The RRI program provided a collaborative environment leading to valuable modifications to the CDSS including prediction explanations, visual aids for survival and integrated education for users producing a user-friendly and quick to use tool.ensuremath</pensuremath>ensuremath<pensuremath>INTERPRETATION: We present a novel, responsibly developed, externally validated AI CDSS trained to predict oesophageal cancer MDT decisions. It represents the foundations of a transformative application of ML, personalised, consistent and efficient MDT decision-support within OC which aligns to RRI principles.ensuremath</pensuremath>ensuremath<pensuremath>FUNDING: Doctoral Studentship for NT (Institute for Life Sciences (University of Southampton) & University Hospital Southampton), UKRI TAS Pump-Priming Grant (TAS_PP_00167).ensuremath</pensuremath>
14.
Le, Thien Doanh; Nguyen, Tan Viet Tuyen; Duy, Tan Le; Ramchurn, Gopal
A multimodal large language model framework for gesture generation in social robots Proceedings Article
In: BEAR Workshop, The 2025 IEEE International Conference on Robot and Human Interactive Communication (25/08/25 - 29/08/25), 2025.
@inproceedings{soton506590,
title = {A multimodal large language model framework for gesture generation in social robots},
author = {Thien Doanh Le and Tan Viet Tuyen Nguyen and Tan Le Duy and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/506590/},
year = {2025},
date = {2025-08-01},
booktitle = {BEAR Workshop, The 2025 IEEE International Conference on Robot and Human Interactive Communication (25/08/25 - 29/08/25)},
abstract = {Non-verbal gestures play a crucial role in social robots, enabling them to signal their intentions to users during human?robot interaction (HRI). While recent research in this domain has primarily focused on robot gesture generation, there remains a limited number of studies on multimodal generation frameworks, where generated gestures are harmonized with other generated modalities, to better convey the robot?s intention to users via a wider range of communication channels. Inspired by recent advancements in multimodal large language models (MLLMs), we propose a novel framework that integrates motion generation models with existing MLLMs to produce high-quality 3D motions without the need for extensive multimodal training. Our framework comprises three key components: a Denoising Diffusion Motion Generation (DDMG) module that maps text descriptions to motion sequences using a diffusion-based approach; a Motion Decoding Alignment (MDA) module that refines motion representations by incorporating signal embeddings generated by an LLM; and a Fusion Module (FM) that integrates motion features trained from previous phases to enhance coherence and realism. We conducted a series of experiments on a publicly available dataset to evaluate the efficiency of the proposed framework in terms of motion quality, diversity, and semantic alignment. The results suggest that our multimodal approach can serve as a powerful controller for robot gesture generation, offering a more scalable and effective solution, particularly for social HRI.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Non-verbal gestures play a crucial role in social robots, enabling them to signal their intentions to users during human?robot interaction (HRI). While recent research in this domain has primarily focused on robot gesture generation, there remains a limited number of studies on multimodal generation frameworks, where generated gestures are harmonized with other generated modalities, to better convey the robot?s intention to users via a wider range of communication channels. Inspired by recent advancements in multimodal large language models (MLLMs), we propose a novel framework that integrates motion generation models with existing MLLMs to produce high-quality 3D motions without the need for extensive multimodal training. Our framework comprises three key components: a Denoising Diffusion Motion Generation (DDMG) module that maps text descriptions to motion sequences using a diffusion-based approach; a Motion Decoding Alignment (MDA) module that refines motion representations by incorporating signal embeddings generated by an LLM; and a Fusion Module (FM) that integrates motion features trained from previous phases to enhance coherence and realism. We conducted a series of experiments on a publicly available dataset to evaluate the efficiency of the proposed framework in terms of motion quality, diversity, and semantic alignment. The results suggest that our multimodal approach can serve as a powerful controller for robot gesture generation, offering a more scalable and effective solution, particularly for social HRI.
15.
Ramchurn, Gopal; Jones, Matt; Owen, Aled Lloyd; Ehsan, Shoaib; Kiden, Sarah; Eke, Damian; McStay, Andrew; Reitmaier, Thomas; Raju, Dani Kalarikalayil; Castenada, Nestor; Sailaja, Neelima; Faith, Becky; Hermann, Antony; Kalra, Kanika; Hawes, Ben; Adams, Rachel; Meier, Patrick; Sharma, Gaurav; Vishwarupe, Varad
Responsible AI. To enable flourishing in and by Low- and Middle- Income Countries (LMICs) Book Section
In: Responsible Ai UK, University of Southampton, 2025.
@incollection{soton506045,
title = {Responsible AI. To enable flourishing in and by Low- and Middle- Income Countries (LMICs)},
author = {Gopal Ramchurn and Matt Jones and Aled Lloyd Owen and Shoaib Ehsan and Sarah Kiden and Damian Eke and Andrew McStay and Thomas Reitmaier and Dani Kalarikalayil Raju and Nestor Castenada and Neelima Sailaja and Becky Faith and Antony Hermann and Kanika Kalra and Ben Hawes and Rachel Adams and Patrick Meier and Gaurav Sharma and Varad Vishwarupe},
url = {https://eprints.soton.ac.uk/506045/},
year = {2025},
date = {2025-07-01},
booktitle = {Responsible Ai UK},
publisher = {University of Southampton},
abstract = {This report explores how governments and industry in Low- and Middle-Income Countries (LMICs) might adopt AI to serve the values, needs and aspirations of individuals and communities, and drive the achievement of development goals. It draws on RAi UK projects working with partners in India and Indonesia and in countries in Africa and in South America, and identifies challenges, early solutions and policy implications. The aim is to identify priority areas and questions for RAi UK and other researchers and organisations to explore further. In this report we focus on the emerging challenges and other outputs of research on AI for LMICs. We convened leaders and members of projects that explore embedding AI tools and resources that are sensitive to local, cultural norms and end-user experience through co-creation with communities.},
keywords = {},
pubstate = {published},
tppubtype = {incollection}
}
This report explores how governments and industry in Low- and Middle-Income Countries (LMICs) might adopt AI to serve the values, needs and aspirations of individuals and communities, and drive the achievement of development goals. It draws on RAi UK projects working with partners in India and Indonesia and in countries in Africa and in South America, and identifies challenges, early solutions and policy implications. The aim is to identify priority areas and questions for RAi UK and other researchers and organisations to explore further. In this report we focus on the emerging challenges and other outputs of research on AI for LMICs. We convened leaders and members of projects that explore embedding AI tools and resources that are sensitive to local, cultural norms and end-user experience through co-creation with communities.
16.
Deshmukh, Jayati; Liang, Zijie; Yazdanpanah, Vahid; Stein, Sebastian; Ramchurn, Savapali D.
Serious games for ethical preference elicitation Proceedings Article
In: AAMAS '25: Proceedings of the 24th International Conference on Autonomous Agents and Multiagent Systems, pp. 3018 – 3020, Association for Computing Machinery, 2025.
@inproceedings{soton498743b,
title = {Serious games for ethical preference elicitation},
author = {Jayati Deshmukh and Zijie Liang and Vahid Yazdanpanah and Sebastian Stein and Savapali D. Ramchurn},
url = {https://eprints.soton.ac.uk/498743/},
year = {2025},
date = {2025-06-01},
booktitle = {AAMAS '25: Proceedings of the 24th International Conference on Autonomous Agents and Multiagent Systems},
pages = {3018 – 3020},
publisher = {Association for Computing Machinery},
abstract = {Autonomous agents acting on behalf of humans must act according to their ethical preferences. However, ethical preferences are latent and abstract and thus it is challenging to elicit them. To address this, we present a serious game that helps elicit ethical preferences in a more dynamic and engaging way than traditional methods such as questionnaires or simple dilemmas.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Autonomous agents acting on behalf of humans must act according to their ethical preferences. However, ethical preferences are latent and abstract and thus it is challenging to elicit them. To address this, we present a serious game that helps elicit ethical preferences in a more dynamic and engaging way than traditional methods such as questionnaires or simple dilemmas.
17.
Abioye, Ayodeji O.; Hunt, William; Gu, Yue; Schneiders, Eike; Naiseh, Mohammad; Archibald, Blair; Sevegnani, Michele; Ramchurn, Sarvapali D.; Fischer, Joel E.; Soorati, Mohammad D.
A user study evaluation of predictive formal modelling at runtime in human-swarm interaction Journal Article
In: ACM Transactions on Human-Robot Interaction, vol. 14, no. 4, 2025.
@article{soton501672,
title = {A user study evaluation of predictive formal modelling at runtime in human-swarm interaction},
author = {Ayodeji O. Abioye and William Hunt and Yue Gu and Eike Schneiders and Mohammad Naiseh and Blair Archibald and Michele Sevegnani and Sarvapali D. Ramchurn and Joel E. Fischer and Mohammad D. Soorati},
url = {https://eprints.soton.ac.uk/501672/},
year = {2025},
date = {2025-06-01},
journal = {ACM Transactions on Human-Robot Interaction},
volume = {14},
number = {4},
abstract = {Formal modelling is often used as part of the design and testing process of software development to ensure that components operate within suitable bounds even in unexpected circumstances. We conducted a user study evaluation of predictive formal modelling (PFM) at runtime in a human-swarm mission to determine the benefit of predictive formal modelling on performance and human-swarm interaction. 180 participants were recruited to perform the role of aerial swarm operators delivering parcels to target locations in a simulation environment. The PFM model was integrated into the simulation software to inform the operator of the estimated mission completion time given the current number of drones deployed. The operator could increase the number of parcels delivered in any time step by adding drones, which also increased costs, thus requiring the use of the minimum number of drones necessary to complete the task in the given time. We collected user feedback using standard survey questionnaires and measured performance using data obtained from the Human And Robot Interactive Swarm (HARIS) simulator. Our results show that PFM increased the performance of the human swarm team without significantly increasing the operators? workload or affecting the system's usability.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Formal modelling is often used as part of the design and testing process of software development to ensure that components operate within suitable bounds even in unexpected circumstances. We conducted a user study evaluation of predictive formal modelling (PFM) at runtime in a human-swarm mission to determine the benefit of predictive formal modelling on performance and human-swarm interaction. 180 participants were recruited to perform the role of aerial swarm operators delivering parcels to target locations in a simulation environment. The PFM model was integrated into the simulation software to inform the operator of the estimated mission completion time given the current number of drones deployed. The operator could increase the number of parcels delivered in any time step by adding drones, which also increased costs, thus requiring the use of the minimum number of drones necessary to complete the task in the given time. We collected user feedback using standard survey questionnaires and measured performance using data obtained from the Human And Robot Interactive Swarm (HARIS) simulator. Our results show that PFM increased the performance of the human swarm team without significantly increasing the operators? workload or affecting the system's usability.
18.
Deshmukh, Jayati; Liang, Zijie; Yazdanpanah, Vahid; Stein, Sebastian; Ramchurn, Sarvalpali D.
Serious games for ethical preference elicitation Proceedings Article
In: AAMAS - 2025 : The 24th International Conference on Autonomous Agents and Multiagent Systems (19/05/25 - 23/05/25), 2025.
@inproceedings{soton498743,
title = {Serious games for ethical preference elicitation},
author = {Jayati Deshmukh and Zijie Liang and Vahid Yazdanpanah and Sebastian Stein and Sarvalpali D. Ramchurn},
url = {https://eprints.soton.ac.uk/498743/},
year = {2025},
date = {2025-05-01},
booktitle = {AAMAS - 2025 : The 24th International Conference on Autonomous Agents and Multiagent Systems (19/05/25 - 23/05/25)},
abstract = {Autonomous agents acting on behalf of humans must act according to their ethical preferences. However, ethical preferences are latent and abstract and thus it is challenging to elicit them. To address this, we present a serious game that helps elicit ethical preferences in a more dynamic and engaging way than traditional methods such as questionnaires or simple dilemmas.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Autonomous agents acting on behalf of humans must act according to their ethical preferences. However, ethical preferences are latent and abstract and thus it is challenging to elicit them. To address this, we present a serious game that helps elicit ethical preferences in a more dynamic and engaging way than traditional methods such as questionnaires or simple dilemmas.
19.
Ismagilov, Timur; Ferrarini, Bruno; Milford, Michael; Tuyen, Nguyen Tan Viet; Ramchurn, Sarvapali D.; Ehsan, Shoaib
On motion blur and deblurring in visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 10, no. 5, pp. 4746–4753, 2025.
@article{soton507348,
title = {On motion blur and deblurring in visual place recognition},
author = {Timur Ismagilov and Bruno Ferrarini and Michael Milford and Nguyen Tan Viet Tuyen and Sarvapali D. Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/507348/},
year = {2025},
date = {2025-03-01},
journal = {IEEE Robotics and Automation Letters},
volume = {10},
number = {5},
pages = {4746–4753},
abstract = {ensuremath<pensuremath>Visual Place Recognition (VPR) in mobile robotics enables robots to localize themselves by recognizing previously visited locations using visual data. While the reliability of VPR methods has been extensively studied under conditions such as changes in illumination, season, weather and viewpoint, the impact of motion blur is relatively unexplored despite its relevance not only in rapid motion scenarios but also in low-light conditions where longer exposure times are necessary. Similarly, the role of image deblurring in enhancing VPR performance under motion blur has received limited attention so far. This letter bridges these gaps by introducing a new benchmark designed to evaluate VPR performance under the influence of motion blur and image deblurring. The benchmark includes three datasets that encompass a wide range of motion blur intensities, providing a comprehensive platform for analysis. Experimental results with several well-established VPR and image deblurring methods provide new insights into the effects of motion blur and the potential improvements achieved through deblurring. Building on these findings, the letter proposes adaptive deblurring strategies for VPR, designed to effectively manage motion blur in dynamic, real-world scenarios.ensuremath</pensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Visual Place Recognition (VPR) in mobile robotics enables robots to localize themselves by recognizing previously visited locations using visual data. While the reliability of VPR methods has been extensively studied under conditions such as changes in illumination, season, weather and viewpoint, the impact of motion blur is relatively unexplored despite its relevance not only in rapid motion scenarios but also in low-light conditions where longer exposure times are necessary. Similarly, the role of image deblurring in enhancing VPR performance under motion blur has received limited attention so far. This letter bridges these gaps by introducing a new benchmark designed to evaluate VPR performance under the influence of motion blur and image deblurring. The benchmark includes three datasets that encompass a wide range of motion blur intensities, providing a comprehensive platform for analysis. Experimental results with several well-established VPR and image deblurring methods provide new insights into the effects of motion blur and the potential improvements achieved through deblurring. Building on these findings, the letter proposes adaptive deblurring strategies for VPR, designed to effectively manage motion blur in dynamic, real-world scenarios.ensuremath</pensuremath>
20.
Abioye, Ayodeji; Hunt, William; Schneiders, Eike; Gu, Yue; Naiseh, Mohammad; Archibald, Blair; Sevegnani, Michele; Ramchurn, Gopal; Fischer, Joel; Soorati, Mohammad
Research Data for Paper: A User Study Evaluation of Predictive Formal Modelling at Runtime in Human-Swarm Interaction Miscellaneous
2025.
@misc{soton500788,
title = {Research Data for Paper: A User Study Evaluation of Predictive Formal Modelling at Runtime in Human-Swarm Interaction},
author = {Ayodeji Abioye and William Hunt and Eike Schneiders and Yue Gu and Mohammad Naiseh and Blair Archibald and Michele Sevegnani and Gopal Ramchurn and Joel Fischer and Mohammad Soorati},
url = {https://eprints.soton.ac.uk/500788/},
year = {2025},
date = {2025-03-01},
publisher = {figshare},
abstract = {ensuremath<span style='white-space:pre-line'ensuremath>This dataset is for the user study conducted to evaluate the performance benefits of predictive formal modelling (PFM) at runtime in a human-swarm interaction experiment. We recruited 180 participants to perform the role of aerial swarm operators conducting a drone delivery mission in a simulation environment using the Human And Robot Interactive Swarm (HARIS) simulator.ensuremath</spanensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
ensuremath<span style='white-space:pre-line'ensuremath>This dataset is for the user study conducted to evaluate the performance benefits of predictive formal modelling (PFM) at runtime in a human-swarm interaction experiment. We recruited 180 participants to perform the role of aerial swarm operators conducting a drone delivery mission in a simulation environment using the Human And Robot Interactive Swarm (HARIS) simulator.ensuremath</spanensuremath>
Selected Publications
Multi-agent signal-less intersection management with dynamic platoon formation
No data available.
https://eprints.soton.ac.uk/478647/
@inproceedings{soton478647,
title = {Multi-agent signal-less intersection management with dynamic platoon formation},
author = {Phuriwat Worrawichaipat and Enrico Gerding and Ioannis Kaparias and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/478647/},
year = {2023},
date = {2023-05-01},
booktitle = {22nd International Conference on Autonomous Agents and Multiagent Systems (29/05/23 - 02/06/23)},
pages = {1542--1550},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
title = {Multi-agent signal-less intersection management with dynamic platoon formation},
author = {Phuriwat Worrawichaipat and Enrico Gerding and Ioannis Kaparias and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/478647/},
year = {2023},
date = {2023-05-01},
booktitle = {22nd International Conference on Autonomous Agents and Multiagent Systems (29/05/23 - 02/06/23)},
pages = {1542--1550},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
AI Foundation Models: initial review, CMA Consultation, TAS Hub Response
No data available.
https://eprints.soton.ac.uk/477553/
@misc{soton477553,
title = {AI Foundation Models: initial review, CMA Consultation, TAS Hub Response},
author = {Joshua Krook and Derek McAuley and Stuart Anderson and John Downer and Peter Winter and Sarvapali D Ramchurn},
url = {https://eprints.soton.ac.uk/477553/},
year = {2023},
date = {2023-06-01},
publisher = {University of Southampton},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
title = {AI Foundation Models: initial review, CMA Consultation, TAS Hub Response},
author = {Joshua Krook and Derek McAuley and Stuart Anderson and John Downer and Peter Winter and Sarvapali D Ramchurn},
url = {https://eprints.soton.ac.uk/477553/},
year = {2023},
date = {2023-06-01},
publisher = {University of Southampton},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
The effect of data visualisation quality and task density on human-swarm interaction
No data available.
https://eprints.soton.ac.uk/479970/
@inproceedings{soton479970,
title = {The effect of data visualisation quality and task density on human-swarm interaction},
author = {Ayodeji Abioye and Mohammad Naiseh and William Hunt and Jediah R Clark and Sarvapali D Ramchurn and Mohammad Soorati},
url = {https://eprints.soton.ac.uk/479970/},
year = {2023},
date = {2023-06-01},
booktitle = {Proceedings of the 2023 32nd IEEE International Conference on Robot and Human Interactive Communication (RO-MAN)},
publisher = {IEEE},
abstract = {Despite the advantages of having robot swarms, human supervision is required for real-world applications. The performance of the human-swarm system depends on several factors including the data availability for the human operators. In this paper, we study the human factors aspect of the human-swarm interaction and investigate how having access to high-quality data can affect the performance of the human-swarm system - the number of tasks completed and the human trust level in operation. We designed an experiment where a human operator is tasked to operate a swarm to identify casualties in an area within a given time period. One group of operators had the option to request high-quality pictures while the other group had to base their decision on the available low-quality images. We performed a user study with 120 participants and recorded their success rate (directly logged via the simulation platform) as well as their workload and trust level (measured through a questionnaire after completing a human-swarm scenario). The findings from our study indicated that the group granted access to high-quality data exhibited an increased workload and placed greater trust in the swarm, thus confirming our initial hypothesis. However, we also found that the number of accurately identified casualties did not significantly vary between the two groups, suggesting that data quality had no impact on the successful completion of tasks.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
title = {The effect of data visualisation quality and task density on human-swarm interaction},
author = {Ayodeji Abioye and Mohammad Naiseh and William Hunt and Jediah R Clark and Sarvapali D Ramchurn and Mohammad Soorati},
url = {https://eprints.soton.ac.uk/479970/},
year = {2023},
date = {2023-06-01},
booktitle = {Proceedings of the 2023 32nd IEEE International Conference on Robot and Human Interactive Communication (RO-MAN)},
publisher = {IEEE},
abstract = {Despite the advantages of having robot swarms, human supervision is required for real-world applications. The performance of the human-swarm system depends on several factors including the data availability for the human operators. In this paper, we study the human factors aspect of the human-swarm interaction and investigate how having access to high-quality data can affect the performance of the human-swarm system - the number of tasks completed and the human trust level in operation. We designed an experiment where a human operator is tasked to operate a swarm to identify casualties in an area within a given time period. One group of operators had the option to request high-quality pictures while the other group had to base their decision on the available low-quality images. We performed a user study with 120 participants and recorded their success rate (directly logged via the simulation platform) as well as their workload and trust level (measured through a questionnaire after completing a human-swarm scenario). The findings from our study indicated that the group granted access to high-quality data exhibited an increased workload and placed greater trust in the swarm, thus confirming our initial hypothesis. However, we also found that the number of accurately identified casualties did not significantly vary between the two groups, suggesting that data quality had no impact on the successful completion of tasks.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Demonstrating performance benefits of human-swarm teaming
Autonomous swarms of robots can bring robustness, scalability and adaptability to safety-critical tasks such as search and rescue but their application is still very limited. Using semi-autonomous swarms with human control can bring robot swarms to real-world applications. Human operators can define goals for the swarm, monitor their performance and interfere with, or overrule, the decisions and behaviour. We present the "Human And Robot Interactive Swarm'' simulator (HARIS) that allows multi-user interaction with a robot swarm and facilitates qualitative and quantitative user studies through simulation of robot swarms completing tasks, from package delivery to search and rescue, with varying levels of human control. In this demonstration, we showcase the simulator by using it to study the performance gain offered by maintaining a "human-in-the-loop'' over a fully autonomous system as an example. This is illustrated in the context of search and rescue, with an autonomous allocation of resources to those in need.
https://eprints.soton.ac.uk/479903/
@inproceedings{soton479903,
title = {Demonstrating performance benefits of human-swarm teaming},
author = {William Hunt and Jack Ryan and Ayodeji O Abioye and Sarvapali D Ramchurn and Mohammad D Soorati},
url = {https://eprints.soton.ac.uk/479903/},
year = {2023},
date = {2023-05-01},
booktitle = {Proceedings of the 2023 International Conference on Autonomous Agents and Multiagent Systems},
pages = {3062--3064},
publisher = {International Foundation for Autonomous Agents and Multiagent Systems (IFAAMAS)},
abstract = {Autonomous swarms of robots can bring robustness, scalability and adaptability to safety-critical tasks such as search and rescue but their application is still very limited. Using semi-autonomous swarms with human control can bring robot swarms to real-world applications. Human operators can define goals for the swarm, monitor their performance and interfere with, or overrule, the decisions and behaviour. We present the "Human And Robot Interactive Swarm'' simulator (HARIS) that allows multi-user interaction with a robot swarm and facilitates qualitative and quantitative user studies through simulation of robot swarms completing tasks, from package delivery to search and rescue, with varying levels of human control. In this demonstration, we showcase the simulator by using it to study the performance gain offered by maintaining a "human-in-the-loop'' over a fully autonomous system as an example. This is illustrated in the context of search and rescue, with an autonomous allocation of resources to those in need.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
title = {Demonstrating performance benefits of human-swarm teaming},
author = {William Hunt and Jack Ryan and Ayodeji O Abioye and Sarvapali D Ramchurn and Mohammad D Soorati},
url = {https://eprints.soton.ac.uk/479903/},
year = {2023},
date = {2023-05-01},
booktitle = {Proceedings of the 2023 International Conference on Autonomous Agents and Multiagent Systems},
pages = {3062--3064},
publisher = {International Foundation for Autonomous Agents and Multiagent Systems (IFAAMAS)},
abstract = {Autonomous swarms of robots can bring robustness, scalability and adaptability to safety-critical tasks such as search and rescue but their application is still very limited. Using semi-autonomous swarms with human control can bring robot swarms to real-world applications. Human operators can define goals for the swarm, monitor their performance and interfere with, or overrule, the decisions and behaviour. We present the "Human And Robot Interactive Swarm'' simulator (HARIS) that allows multi-user interaction with a robot swarm and facilitates qualitative and quantitative user studies through simulation of robot swarms completing tasks, from package delivery to search and rescue, with varying levels of human control. In this demonstration, we showcase the simulator by using it to study the performance gain offered by maintaining a "human-in-the-loop'' over a fully autonomous system as an example. This is illustrated in the context of search and rescue, with an autonomous allocation of resources to those in need.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
2026
Deshmukh, Jayati; Yazdanpanah, Vahid; Stein, Sebastian; Ramchurn, Gopal
The triad of identity, trust and responsibility in multi-agent systems Proceedings Article
In: 25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26), 2026.
@inproceedings{soton509930,
title = {The triad of identity, trust and responsibility in multi-agent systems},
author = {Jayati Deshmukh and Vahid Yazdanpanah and Sebastian Stein and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/509930/},
year = {2026},
date = {2026-05-01},
booktitle = {25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26)},
abstract = {The design of autonomous AI agents that behave responsibly and foster trust in open multi-agent systems remains a fundamental challenge. Traditional game-theoretical approaches largely assume self-interested behaviour, yet real-world collaborations among humans often rely on prosocial considerations that extend beyond individual utility. To address this, for the first time in this paper, we investigate the triad of identity, responsibility, and trust as core elements shaping responsible multi-agent behaviour. We propose a novel agent model, building on the notion of Computational Transcendence, which equips agents with an elastic sense of identity, enabling them to incorporate the welfare of others into their decision-making. Our framework integrates subjective (identity-based) and objective (experience-based and reputation-based) components of trust. Using Iterated Prisoner?s Dilemma (IPD) simulations on different network structures, we analyse how varying levels of identity and trust affect responsible behaviour. Results demonstrate that the interplay of these three concepts can promote emergent responsibility, mitigate exploitation, and sustain long-term cooperation in dynamic multi-agent environments. We argue that this triadic perspective provides a principled foundation for designing trustworthy, responsible, and identity/value aware agents with implications for future human?AI collaboration.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The design of autonomous AI agents that behave responsibly and foster trust in open multi-agent systems remains a fundamental challenge. Traditional game-theoretical approaches largely assume self-interested behaviour, yet real-world collaborations among humans often rely on prosocial considerations that extend beyond individual utility. To address this, for the first time in this paper, we investigate the triad of identity, responsibility, and trust as core elements shaping responsible multi-agent behaviour. We propose a novel agent model, building on the notion of Computational Transcendence, which equips agents with an elastic sense of identity, enabling them to incorporate the welfare of others into their decision-making. Our framework integrates subjective (identity-based) and objective (experience-based and reputation-based) components of trust. Using Iterated Prisoner?s Dilemma (IPD) simulations on different network structures, we analyse how varying levels of identity and trust affect responsible behaviour. Results demonstrate that the interplay of these three concepts can promote emergent responsibility, mitigate exploitation, and sustain long-term cooperation in dynamic multi-agent environments. We argue that this triadic perspective provides a principled foundation for designing trustworthy, responsible, and identity/value aware agents with implications for future human?AI collaboration.
Georgara, Athina; Deshmukh, Jayati; Ramchurn, Gopal
GreenLine: A delay-tolerant mechanism design for grid capacity allocation Proceedings Article
In: 25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26), pp. 3323–3325, 2026.
@inproceedings{soton511473,
title = {GreenLine: A delay-tolerant mechanism design for grid capacity allocation},
author = {Athina Georgara and Jayati Deshmukh and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/511473/},
year = {2026},
date = {2026-05-01},
booktitle = {25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26)},
pages = {3323–3325},
abstract = {This work presents GreenLine, a new auction-based mechanism for integrating renewable power plants (RPP) in the power grid. We address a challenging task, where RPP owners want to maximise their profits by installing new power plants in the grid; while, the Distribution Network Operator (DNO) seeks to maximise the generated power while reducing potential installation delays due to growing power demand. This paper formulates this task as a multi-agent system, studies its useful properties such as incentive compatibility, individual rationality and economical efficiency, and discusses GreenLine under different deployment variations.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This work presents GreenLine, a new auction-based mechanism for integrating renewable power plants (RPP) in the power grid. We address a challenging task, where RPP owners want to maximise their profits by installing new power plants in the grid; while, the Distribution Network Operator (DNO) seeks to maximise the generated power while reducing potential installation delays due to growing power demand. This paper formulates this task as a multi-agent system, studies its useful properties such as incentive compatibility, individual rationality and economical efficiency, and discusses GreenLine under different deployment variations.
Singh, Lokesh; Deshmukh, Jayati; Georgara, Athina; Nguyen, Tan Viet Tuyen; Ramchurn, Gopal
CareOps: A multi agent control room for independent living with care Proceedings Article
In: 25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26), pp. 4146–4148, 2026.
@inproceedings{soton511472,
title = {CareOps: A multi agent control room for independent living with care},
author = {Lokesh Singh and Jayati Deshmukh and Athina Georgara and Tan Viet Tuyen Nguyen and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/511472/},
year = {2026},
date = {2026-05-01},
booktitle = {25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026) (25/05/26 - 29/05/26)},
pages = {4146–4148},
abstract = {This paper introduces CareOps, a multi-agent control-room dashboard for Independent Living with Care (ILWC) that coordinates heterogeneous sensors and a Buddy robot across multiple simulated homes. Agents fuse radar, audio, bed, passive infrared (PIR), and gait data into single prioritised incidents with human-readable explanations, and support one-click dispatch of either a robot (via socket) or a human carer (via smartphone notification). The demo presents a single decision-support workflow that orchestrates sensors, software agents, and robots through a unified interface. It provides a human-in-the-loop platform for exploring how professional care staff understand multi-sensor evidence and choose between robot and human responses.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper introduces CareOps, a multi-agent control-room dashboard for Independent Living with Care (ILWC) that coordinates heterogeneous sensors and a Buddy robot across multiple simulated homes. Agents fuse radar, audio, bed, passive infrared (PIR), and gait data into single prioritised incidents with human-readable explanations, and support one-click dispatch of either a robot (via socket) or a human carer (via smartphone notification). The demo presents a single decision-support workflow that orchestrates sensors, software agents, and robots through a unified interface. It provides a human-in-the-loop platform for exploring how professional care staff understand multi-sensor evidence and choose between robot and human responses.
Hafez, Muhammad Burhan; Miller, Emily; Milford, Michael J.; Ramchurn, Gopal; Ehsan, Shoaib
Through the lens of doubt: robust and efficient uncertainty estimation for visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 11, no. 5, pp. 5899–5906, 2026.
@article{soton510726,
title = {Through the lens of doubt: robust and efficient uncertainty estimation for visual place recognition},
author = {Muhammad Burhan Hafez and Emily Miller and Michael J. Milford and Gopal Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/510726/},
year = {2026},
date = {2026-03-01},
journal = {IEEE Robotics and Automation Letters},
volume = {11},
number = {5},
pages = {5899–5906},
abstract = {ensuremath<pensuremath>Visual Place Recognition (VPR) enables robots and autonomous vehicles to identify previously visited locations by matching current observations against a database of known places. However, VPR systems face significant challenges when deployed across varying visual environments, lighting conditions, seasonal changes, and viewpoints changes. Failure-critical VPR applications, such as loop closure detection in simultaneous localization and mapping (SLAM) pipelines, require robust estimation of place matching uncertainty. We propose three training-free uncertainty metrics that estimate prediction confidence by analyzing inherent statistical patterns in similarity scores from any existing VPR method. Similarity Distribution (SD) quantifies match distinctiveness by measuring score separation between candidates; Ratio Spread (RS) evaluates competitive ambiguity among top-scoring locations; and Statistical Uncertainty (SU) is a combination of SD and RS that provides a unified metric that generalizes across datasets and VPR methods without requiring validation data to select the optimal metric. All three metrics operate without additional model training, architectural modifications, or computationally expensive geometric verification. Comprehensive evaluation across nine state-of-the-art VPR methods and six benchmark datasets confirms that our metrics excel at discriminating between correct and incorrect VPR matches, and consistently outperform existing approaches while maintaining negligible computational overhead, making it deployable for real-time robotic applications across varied environmental conditions with improved precision-recall performance.ensuremath</pensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Visual Place Recognition (VPR) enables robots and autonomous vehicles to identify previously visited locations by matching current observations against a database of known places. However, VPR systems face significant challenges when deployed across varying visual environments, lighting conditions, seasonal changes, and viewpoints changes. Failure-critical VPR applications, such as loop closure detection in simultaneous localization and mapping (SLAM) pipelines, require robust estimation of place matching uncertainty. We propose three training-free uncertainty metrics that estimate prediction confidence by analyzing inherent statistical patterns in similarity scores from any existing VPR method. Similarity Distribution (SD) quantifies match distinctiveness by measuring score separation between candidates; Ratio Spread (RS) evaluates competitive ambiguity among top-scoring locations; and Statistical Uncertainty (SU) is a combination of SD and RS that provides a unified metric that generalizes across datasets and VPR methods without requiring validation data to select the optimal metric. All three metrics operate without additional model training, architectural modifications, or computationally expensive geometric verification. Comprehensive evaluation across nine state-of-the-art VPR methods and six benchmark datasets confirms that our metrics excel at discriminating between correct and incorrect VPR matches, and consistently outperform existing approaches while maintaining negligible computational overhead, making it deployable for real-time robotic applications across varied environmental conditions with improved precision-recall performance.ensuremath</pensuremath>
Tuyen, Nguyen Tan Viet; Georgara, Athina; Singh, Lokesh; Deshmukh, Jayati; Davey, Sean; Tisdale, Paul N.; Ramchurn, Sarvapali
Supporting ?Stop and Watch? in elderly care with socially assistive robots: insights from a participatory design workshop Proceedings Article
In: Baillie, Lynne; Smart, William D.; Graaf, Maartje De; Gombolay, Matthew; Torre, Ilaria (Ed.): Companion Proceedings of the 21st ACM/IEEE International Conference on Human-Robot Interaction, HRI Companion 2026, pp. 302–306, ACM Press, 2026.
@inproceedings{soton511593,
title = {Supporting ?Stop and Watch? in elderly care with socially assistive robots: insights from a participatory design workshop},
author = {Nguyen Tan Viet Tuyen and Athina Georgara and Lokesh Singh and Jayati Deshmukh and Sean Davey and Paul N. Tisdale and Sarvapali Ramchurn},
editor = {Lynne Baillie and William D. Smart and Maartje De Graaf and Matthew Gombolay and Ilaria Torre},
url = {https://eprints.soton.ac.uk/511593/},
year = {2026},
date = {2026-03-01},
urldate = {2026-03-01},
booktitle = {Companion Proceedings of the 21st ACM/IEEE International Conference on Human-Robot Interaction, HRI Companion 2026},
pages = {302–306},
publisher = {ACM Press},
abstract = {ensuremath<pensuremath>?Stop and Watch? is an early-warning tool adopted by the UK NHS and is widely used in elderly care settings. The tool helps caregivers to recognise abnormal changes in residents? health. Despite its clinical value, the process remains highly manual, workload-intensive, and vulnerable to missed observations, particularly in environments facing staff shortages, frequent staff rotations, and increasing care demands. We argue that AI-based systems such as Socially Assistive Robots (SARs) and Assistive Living Technologies (ALTs) offer promising avenues for supporting and enhancing the ?Stop and Watch? tool. However, designing such systems requires a multidisciplinary effort to establish a comprehensive understanding of current practices and the priorities and concerns of all relevant stakeholders. This paper presents insights from a participatory design workshop held in a care home in the UK to explore how SARs and ALTs could meaningfully support the ?Stop and Watch? tool, understand stakeholders? expectations, perceived benefits, and concerns regarding deployment in this sensitive context.ensuremath</pensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
ensuremath<pensuremath>?Stop and Watch? is an early-warning tool adopted by the UK NHS and is widely used in elderly care settings. The tool helps caregivers to recognise abnormal changes in residents? health. Despite its clinical value, the process remains highly manual, workload-intensive, and vulnerable to missed observations, particularly in environments facing staff shortages, frequent staff rotations, and increasing care demands. We argue that AI-based systems such as Socially Assistive Robots (SARs) and Assistive Living Technologies (ALTs) offer promising avenues for supporting and enhancing the ?Stop and Watch? tool. However, designing such systems requires a multidisciplinary effort to establish a comprehensive understanding of current practices and the priorities and concerns of all relevant stakeholders. This paper presents insights from a participatory design workshop held in a care home in the UK to explore how SARs and ALTs could meaningfully support the ?Stop and Watch? tool, understand stakeholders? expectations, perceived benefits, and concerns regarding deployment in this sensitive context.ensuremath</pensuremath>
Everett, Gregory Alan
Models and algorithms to optimise team performance in football PhD Thesis
University of Southampton, 2026.
@phdthesis{soton509865,
title = {Models and algorithms to optimise team performance in football},
author = {Gregory Alan Everett},
url = {https://eprints.soton.ac.uk/509865/},
year = {2026},
date = {2026-03-01},
publisher = {University of Southampton},
school = {University of Southampton},
abstract = {With the growing use of Artificial Intelligence (AI) and machine learning, there is increasing potential to model and optimise team behaviour across important domains such as disaster response, security, and team sports. These domains are inherently spatiotemporal, requiring models that capture both spatial and temporal dynamics. This thesis focuses on football, a complex, dynamic sport with rich spatiotemporal data and clear objectives, making it an ideal testbed for developing and validating team-based AI models. Moreover, football analytics is a rapidly expanding industry, with European clubs generating ?38 billion in revenue during the 2023/24 season, driving the demand for models that provide competitive advantages.ensuremath<br/ensuremath>ensuremath<br/ensuremath>This thesis proposes a number of novel methods that utilise spatiotemporal data to advance team prediction, analysis, and decision-making in football and other team-based domains. In particular, we introduce a spatiotemporal agent behaviour imputation model that reduces predictive error by 62% compared to baselines in limited observability settings, significantly improving the accessibility of off-ball football analytics through imputed tracking data. We also present a novel spatial teamwork model, combining Monte Carlo tree search (MCTS) and linear programming, that optimises agent decision-making in real-world football defence, reducing opponent threat by 24%. In addition, we develop new metrics, derived from a graph attention network (GAT), to assign credit to indirect agent contributions in team-based defence. The GAT model predicts football passes with a 6% reduction in loss compared to baselines, and we show how these new metrics can greatly improve off-ball football player evaluation. Finally, we propose a dynamic team formation and agent replacement model that accounts for agent fatigue and unavailability and optimises decision-making using a multi-step MCTS algorithm. Applied to football team selection and substitutions, this model improves long-term team performance by 1% and reduces first-team injuries by 15%. This thesis also highlights key remaining challenges for AI in football, including the use of richer player data (e.g., body pose) and greater use of explainable models to build trust in clubs.},
keywords = {},
pubstate = {published},
tppubtype = {phdthesis}
}
With the growing use of Artificial Intelligence (AI) and machine learning, there is increasing potential to model and optimise team behaviour across important domains such as disaster response, security, and team sports. These domains are inherently spatiotemporal, requiring models that capture both spatial and temporal dynamics. This thesis focuses on football, a complex, dynamic sport with rich spatiotemporal data and clear objectives, making it an ideal testbed for developing and validating team-based AI models. Moreover, football analytics is a rapidly expanding industry, with European clubs generating ?38 billion in revenue during the 2023/24 season, driving the demand for models that provide competitive advantages.ensuremath<br/ensuremath>ensuremath<br/ensuremath>This thesis proposes a number of novel methods that utilise spatiotemporal data to advance team prediction, analysis, and decision-making in football and other team-based domains. In particular, we introduce a spatiotemporal agent behaviour imputation model that reduces predictive error by 62% compared to baselines in limited observability settings, significantly improving the accessibility of off-ball football analytics through imputed tracking data. We also present a novel spatial teamwork model, combining Monte Carlo tree search (MCTS) and linear programming, that optimises agent decision-making in real-world football defence, reducing opponent threat by 24%. In addition, we develop new metrics, derived from a graph attention network (GAT), to assign credit to indirect agent contributions in team-based defence. The GAT model predicts football passes with a 6% reduction in loss compared to baselines, and we show how these new metrics can greatly improve off-ball football player evaluation. Finally, we propose a dynamic team formation and agent replacement model that accounts for agent fatigue and unavailability and optimises decision-making using a multi-step MCTS algorithm. Applied to football team selection and substitutions, this model improves long-term team performance by 1% and reduces first-team injuries by 15%. This thesis also highlights key remaining challenges for AI in football, including the use of richer player data (e.g., body pose) and greater use of explainable models to build trust in clubs.
Ramchurn, Gopal; Neff, Gina; Parisio, Isabela; Kiden, Sarah; Georgara, Athina; Stumpf, Simone; Wong, Mark; Shahandashti, Siamak F.; Aristodemou, Marios
Responsible AI UK response to the DSIT?s call for evidence on the AI Growth Lab Technical Report
no. 10.5258/RAi/005, 2026.
@techreport{soton508020,
title = {Responsible AI UK response to the DSIT?s call for evidence on the AI Growth Lab},
author = {Gopal Ramchurn and Gina Neff and Isabela Parisio and Sarah Kiden and Athina Georgara and Simone Stumpf and Mark Wong and Siamak F. Shahandashti and Marios Aristodemou},
url = {https://eprints.soton.ac.uk/508020/},
year = {2026},
date = {2026-01-01},
number = {10.5258/RAi/005},
publisher = {University of Southampton},
abstract = {A response to the Department for Science, Innovation and Technology (DSIT) open call for evidence regarding the AI Growth Lab1 on behalf of Responsible AI UK (RAi UK), an open and multidisciplinary network that brings together experts from across the four nations of the UK to understand how we should shape the development of AI to benefit people, communities and society},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
A response to the Department for Science, Innovation and Technology (DSIT) open call for evidence regarding the AI Growth Lab1 on behalf of Responsible AI UK (RAi UK), an open and multidisciplinary network that brings together experts from across the four nations of the UK to understand how we should shape the development of AI to benefit people, communities and society
2025
Reyes-Cruz, Gisela; Kiden, Sarah; Azim, Tayyaba; Bergin, Aislinn Gomez; Choi, Sena; Eke, Damian; Klyshbekova, Maira; Peter, Oriane; Waheed, Maria; Devlin, Kate; Fischer, Joel; Vallejos, Elvira Perez; Ramchurn, Gopal; Stein, Sebastian
Response to call for input for EMRTD study ?Artificial Intelligence, Cultural Rights, and the Right to Development? Technical Report
no. 10.25878/7nmj-0t79, 2025.
@techreport{soton511244,
title = {Response to call for input for EMRTD study ?Artificial Intelligence, Cultural Rights, and the Right to Development?},
author = {Gisela Reyes-Cruz and Sarah Kiden and Tayyaba Azim and Aislinn Gomez Bergin and Sena Choi and Damian Eke and Maira Klyshbekova and Oriane Peter and Maria Waheed and Kate Devlin and Joel Fischer and Elvira Perez Vallejos and Gopal Ramchurn and Sebastian Stein},
url = {https://eprints.soton.ac.uk/511244/},
year = {2025},
date = {2025-11-01},
number = {10.25878/7nmj-0t79},
publisher = {University of Nottingham},
abstract = {Responsible AI UK's response to the Call for Input for EMRTD study "Artificial Intelligence, Cultural Rights, and the Right to Development" issued by the Expert Mechanism on the Right to Development United Nations Human Rights Office of the High Commissioner (OHCHR).ensuremath<br/ensuremath>ensuremath<br/ensuremath>https://www.ohchr.org/en/calls-for-input/2025/call-input-emrtd-study-artificial-intelligence-cultural-rights-and-right},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
Responsible AI UK's response to the Call for Input for EMRTD study "Artificial Intelligence, Cultural Rights, and the Right to Development" issued by the Expert Mechanism on the Right to Development United Nations Human Rights Office of the High Commissioner (OHCHR).ensuremath<br/ensuremath>ensuremath<br/ensuremath>https://www.ohchr.org/en/calls-for-input/2025/call-input-emrtd-study-artificial-intelligence-cultural-rights-and-right
Owen, Aled Lloyd; Ramchurn, Gopal; Dasgupta, Prokar; Ao, Shuang; Barnard, Pepita; Deshmukh, Jayati; Parisio, Isabela; Singh, Lokesh; Valoor, Adarsh; Waheed, Maria; Braunschweig, Bertrand; Quintero, Karla; Poretschkin, Maximilian; Hawes, Ben; Khomh, Foutse
Advancing Trustworthy Artificial Intelligence: Lessons Learned and Emerging Challenges Proceedings Article
In: Responsible Ai, University of Southampton, 2025.
@inproceedings{soton506660,
title = {Advancing Trustworthy Artificial Intelligence: Lessons Learned and Emerging Challenges},
author = {Aled Lloyd Owen and Gopal Ramchurn and Prokar Dasgupta and Shuang Ao and Pepita Barnard and Jayati Deshmukh and Isabela Parisio and Lokesh Singh and Adarsh Valoor and Maria Waheed and Bertrand Braunschweig and Karla Quintero and Maximilian Poretschkin and Ben Hawes and Foutse Khomh},
url = {https://eprints.soton.ac.uk/506660/},
year = {2025},
date = {2025-11-01},
booktitle = {Responsible Ai},
publisher = {University of Southampton},
abstract = {Government white paper},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Government white paper
Kiden, Sarah; Peter, Oriane; Reyes-Cruz, Gisela; Klyshbekova, Maira; Choi, Sena; Bergin, Aislinn Gomez; Waheed, Maria; Eke, Damian; Azim, Tayyaba; Ramchurn, Sarvapali; Stein, Sebastian; Vallejos, Elvira Perez; Devlin, Kate; Fischer, Joel E.
Back to the communities: a mixed-methods and community-driven evaluation of cultural sensitivity in text-to-image models Miscellaneous
2025.
@misc{soton507303,
title = {Back to the communities: a mixed-methods and community-driven evaluation of cultural sensitivity in text-to-image models},
author = {Sarah Kiden and Oriane Peter and Gisela Reyes-Cruz and Maira Klyshbekova and Sena Choi and Aislinn Gomez Bergin and Maria Waheed and Damian Eke and Tayyaba Azim and Sarvapali Ramchurn and Sebastian Stein and Elvira Perez Vallejos and Kate Devlin and Joel E. Fischer},
url = {https://eprints.soton.ac.uk/507303/},
year = {2025},
date = {2025-10-01},
publisher = {arXiv},
abstract = {Evidence shows that text-to-image (T2I) models disproportionately reflect Western cultural norms, amplifying misrepresentation and harms to minority groups. However, evaluating cultural sensitivity is inherently complex due to its fluid and multifaceted nature. This paper draws on a state-of-the-art review and co-creation workshops involving 59 individuals from 19 different countries. We developed and validated a mixed-methods community-based evaluation methodology to assess cultural sensitivity in T2I models, which embraces first-person methods. Quantitative scores and qualitative inquiries expose convergence and disagreement within and across communities, illuminate the downstream consequences of misrepresentation, and trace how training data shaped by unequal power relations distort depictions. Extensive assessments are constrained by high resource requirements and the dynamic nature of culture, a tension we alleviate through a context-based and iterative methodology. The paper provides actionable recommendations for stakeholders, highlighting pathways to investigate the sources, mechanisms, and impacts of cultural (mis)representation in T2I models.},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Evidence shows that text-to-image (T2I) models disproportionately reflect Western cultural norms, amplifying misrepresentation and harms to minority groups. However, evaluating cultural sensitivity is inherently complex due to its fluid and multifaceted nature. This paper draws on a state-of-the-art review and co-creation workshops involving 59 individuals from 19 different countries. We developed and validated a mixed-methods community-based evaluation methodology to assess cultural sensitivity in T2I models, which embraces first-person methods. Quantitative scores and qualitative inquiries expose convergence and disagreement within and across communities, illuminate the downstream consequences of misrepresentation, and trace how training data shaped by unequal power relations distort depictions. Extensive assessments are constrained by high resource requirements and the dynamic nature of culture, a tension we alleviate through a context-based and iterative methodology. The paper provides actionable recommendations for stakeholders, highlighting pathways to investigate the sources, mechanisms, and impacts of cultural (mis)representation in T2I models.
Ramchurn, Gopal; Owen, Aled Lloyd; Ao, Shuang; Barnard, Pepita; Bergin, Aislinn Gomez; Parisio, Isabela; Valoor, Adarsh; Waheed, Maria; Holter, Carolyn Ten; Portillo, Virginia; Procter, Rob; Batterham, Paul; Downer, John; Winter, Peter; Krook, Joshua; Blockx, Jan; Braunschweig, Bertrand; Quintero, Karla; Poretschkin, Maximilian; Becker, Daniel; Campbell-Ratcliffe, Emily; Hawes, Ben; Shaw, Patricia; Patel, Reema; Tewari, Ashish; Thomas, Alec; Duncan, Paul; Gillings, Eliot
Frameworks and Toolkits for Assuring Responsible AI Book Section
In: Responsible Ai UK, University of Southampton, 2025.
@incollection{soton506057,
title = {Frameworks and Toolkits for Assuring Responsible AI},
author = {Gopal Ramchurn and Aled Lloyd Owen and Shuang Ao and Pepita Barnard and Aislinn Gomez Bergin and Isabela Parisio and Adarsh Valoor and Maria Waheed and Carolyn Ten Holter and Virginia Portillo and Rob Procter and Paul Batterham and John Downer and Peter Winter and Joshua Krook and Jan Blockx and Bertrand Braunschweig and Karla Quintero and Maximilian Poretschkin and Daniel Becker and Emily Campbell-Ratcliffe and Ben Hawes and Patricia Shaw and Reema Patel and Ashish Tewari and Alec Thomas and Paul Duncan and Eliot Gillings},
url = {https://eprints.soton.ac.uk/506057/},
year = {2025},
date = {2025-10-01},
urldate = {2025-10-01},
booktitle = {Responsible Ai UK},
publisher = {University of Southampton},
abstract = {This report explores experiences of using tools for assuring responsible practice in the build, deployment, and governance of AI systems, in particular frameworks and toolkits developed and published by RAi UK-funded projects and by other organisations. We examine evidence of how and why these tools are used, and of their usefulness and their limitations. We look for lessons that could improve AI assurance in the future, including potentially using AI-based tools to support AI assurance. The aim is to identify priority areas and key questions for RAi UK and other researchers and organisations to explore further, with the goal of improving the fit between the supply of, and demand for, tools that support organisational assurance of responsible AI .Responsible Ai UK and our partner for this event, Confiance.ai, brought people together from across government, public services, and industry, as well as project teams that develop support for responsible AI. The aim was to clarify how people should approach the development and use of toolkits and/or frameworks in practice. We wanted to synthesise findings from our research across the programme and to find and address any gaps. The workshop took place online on Thursday 10 April 2025.},
keywords = {},
pubstate = {published},
tppubtype = {incollection}
}
This report explores experiences of using tools for assuring responsible practice in the build, deployment, and governance of AI systems, in particular frameworks and toolkits developed and published by RAi UK-funded projects and by other organisations. We examine evidence of how and why these tools are used, and of their usefulness and their limitations. We look for lessons that could improve AI assurance in the future, including potentially using AI-based tools to support AI assurance. The aim is to identify priority areas and key questions for RAi UK and other researchers and organisations to explore further, with the goal of improving the fit between the supply of, and demand for, tools that support organisational assurance of responsible AI .Responsible Ai UK and our partner for this event, Confiance.ai, brought people together from across government, public services, and industry, as well as project teams that develop support for responsible AI. The aim was to clarify how people should approach the development and use of toolkits and/or frameworks in practice. We wanted to synthesise findings from our research across the programme and to find and address any gaps. The workshop took place online on Thursday 10 April 2025.
Thavanesan, Navamayooran; Naiseh, Mohammad; Terol, Miguel; Rahman, Saqib Andrew; Hill, Samuel Luke; Parfitt, Charlotte; Walters, Zoë S; Ramchurn, Sarvapali; Markar, Sheraz; Owen, Richard; Maynard, Nick; Azim, Tayyaba; Belkatir, Zehor; Perez, Elvira Vallejos; McCord, Mimi; Underwood, Tim; Vigneswaran, Ganesh
The oesophageal cancer multi-disciplinary tool: a co-designed, externally validated, machine learning clinical decision support system Journal Article
In: EClinicalMedicine, vol. 89, pp. 103527, 2025, (For the purposes of open access, the authors have applied a Creative Commons attribution license (CC-BY) to any Author Accepted Manuscript version arising from this submission.).
@article{soton506614,
title = {The oesophageal cancer multi-disciplinary tool: a co-designed, externally validated, machine learning clinical decision support system},
author = {Navamayooran Thavanesan and Mohammad Naiseh and Miguel Terol and Saqib Andrew Rahman and Samuel Luke Hill and Charlotte Parfitt and Zoë S Walters and Sarvapali Ramchurn and Sheraz Markar and Richard Owen and Nick Maynard and Tayyaba Azim and Zehor Belkatir and Elvira Vallejos Perez and Mimi McCord and Tim Underwood and Ganesh Vigneswaran},
url = {https://eprints.soton.ac.uk/506614/},
year = {2025},
date = {2025-09-01},
journal = {EClinicalMedicine},
volume = {89},
pages = {103527},
abstract = {ensuremath<pensuremath>BACKGROUND: The oesophageal cancer (OC) multi-disciplinary team (MDT) operates under significant pressures, handling complex decision-making. Machine learning (ML) can learn complex decision-making paradigms to improve efficiency, consistency, and cost if trained and deployed responsibly. We present an externally validated ML-based clinical decision support system (CDSS) designed to predict OC MDT treatment decisions and prognosticate palliative scenarios, co-designed using Responsible Research and Innovation (RRI) principles.ensuremath</pensuremath>ensuremath<pensuremath>METHODS: Clinicopathological data collected from 1931 patients between 4th September 2009, and 8th November 2022 were used to test and validate models trained through four ML algorithms to predict curative and palliative treatment pathways along with palliative prognosis. 953 OC cases treated at University Hospitals Southampton (UHS) were used to train ML models which were externally validated on 978 OC cases from Oxford University Hospitals (OUH). Model performance was evaluated using Area Under Curve (AUC) for treatment classifiers and calibration curves for survival models. A parallel RRI program at the University of Southampton (United Kingdom) combining clinician interviews and inter-disciplinary workshops was conducted between 16.3.23 and 23.5.24. The RRI program comprised a group of 17 domain experts comprising programmers, computer scientists, clinicians and patient representatives to allow end-users to contribute towards the co-design of the CDSS user interface.ensuremath</pensuremath>ensuremath<pensuremath>FINDINGS: Cohorts differed in baseline characteristics, with the external cohort (OUH) being younger, having better performance status, and a higher prevalence of pulmonary and vascular disease. Despite these differences, on internal validation (UHS cohort) mean AUCs for the primary treatment model were: MLR 0.905 $±$ 0.048, XGB 0.909 $±$ 0.044 and RF 0.883 $±$ 0.059 (k = 5 cross-validation) and MLR 0.866 (95% CI 0.866-0.867), XGB 0.863 (0.862-0.864), RF 0.863 (0.867-0.868) on bootstrapped resampling. For the palliative classifier, mean AUCs were: MLR 0.805 $±$ 0.096, XGB 0.815 $±$ 0.081 and RF 0.793 $±$ 0.083 (k = 5 cross-validation) and MLR 0.736 (95% CI 0.734-0.737), XGB 0.799 (0.798-0.800), RF 0.781 (0.778-0.782) on bootstrapped resampling. On external validation (OUH cohort), AUCs were MLR 0.894, XGB 0.887 and RF 0.891 for the primary treatment model and MLR 0.711, XGB 0.742 and RF 0.730 for the palliative treatment classifier. Predicted survival probability from the palliative survival model was well calibrated over the first 12 months post-diagnosis in both cohorts. The RRI program provided a collaborative environment leading to valuable modifications to the CDSS including prediction explanations, visual aids for survival and integrated education for users producing a user-friendly and quick to use tool.ensuremath</pensuremath>ensuremath<pensuremath>INTERPRETATION: We present a novel, responsibly developed, externally validated AI CDSS trained to predict oesophageal cancer MDT decisions. It represents the foundations of a transformative application of ML, personalised, consistent and efficient MDT decision-support within OC which aligns to RRI principles.ensuremath</pensuremath>ensuremath<pensuremath>FUNDING: Doctoral Studentship for NT (Institute for Life Sciences (University of Southampton) & University Hospital Southampton), UKRI TAS Pump-Priming Grant (TAS_PP_00167).ensuremath</pensuremath>},
note = {For the purposes of open access, the authors have applied a Creative Commons attribution license (CC-BY) to any Author Accepted Manuscript version arising from this submission.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>BACKGROUND: The oesophageal cancer (OC) multi-disciplinary team (MDT) operates under significant pressures, handling complex decision-making. Machine learning (ML) can learn complex decision-making paradigms to improve efficiency, consistency, and cost if trained and deployed responsibly. We present an externally validated ML-based clinical decision support system (CDSS) designed to predict OC MDT treatment decisions and prognosticate palliative scenarios, co-designed using Responsible Research and Innovation (RRI) principles.ensuremath</pensuremath>ensuremath<pensuremath>METHODS: Clinicopathological data collected from 1931 patients between 4th September 2009, and 8th November 2022 were used to test and validate models trained through four ML algorithms to predict curative and palliative treatment pathways along with palliative prognosis. 953 OC cases treated at University Hospitals Southampton (UHS) were used to train ML models which were externally validated on 978 OC cases from Oxford University Hospitals (OUH). Model performance was evaluated using Area Under Curve (AUC) for treatment classifiers and calibration curves for survival models. A parallel RRI program at the University of Southampton (United Kingdom) combining clinician interviews and inter-disciplinary workshops was conducted between 16.3.23 and 23.5.24. The RRI program comprised a group of 17 domain experts comprising programmers, computer scientists, clinicians and patient representatives to allow end-users to contribute towards the co-design of the CDSS user interface.ensuremath</pensuremath>ensuremath<pensuremath>FINDINGS: Cohorts differed in baseline characteristics, with the external cohort (OUH) being younger, having better performance status, and a higher prevalence of pulmonary and vascular disease. Despite these differences, on internal validation (UHS cohort) mean AUCs for the primary treatment model were: MLR 0.905 $±$ 0.048, XGB 0.909 $±$ 0.044 and RF 0.883 $±$ 0.059 (k = 5 cross-validation) and MLR 0.866 (95% CI 0.866-0.867), XGB 0.863 (0.862-0.864), RF 0.863 (0.867-0.868) on bootstrapped resampling. For the palliative classifier, mean AUCs were: MLR 0.805 $±$ 0.096, XGB 0.815 $±$ 0.081 and RF 0.793 $±$ 0.083 (k = 5 cross-validation) and MLR 0.736 (95% CI 0.734-0.737), XGB 0.799 (0.798-0.800), RF 0.781 (0.778-0.782) on bootstrapped resampling. On external validation (OUH cohort), AUCs were MLR 0.894, XGB 0.887 and RF 0.891 for the primary treatment model and MLR 0.711, XGB 0.742 and RF 0.730 for the palliative treatment classifier. Predicted survival probability from the palliative survival model was well calibrated over the first 12 months post-diagnosis in both cohorts. The RRI program provided a collaborative environment leading to valuable modifications to the CDSS including prediction explanations, visual aids for survival and integrated education for users producing a user-friendly and quick to use tool.ensuremath</pensuremath>ensuremath<pensuremath>INTERPRETATION: We present a novel, responsibly developed, externally validated AI CDSS trained to predict oesophageal cancer MDT decisions. It represents the foundations of a transformative application of ML, personalised, consistent and efficient MDT decision-support within OC which aligns to RRI principles.ensuremath</pensuremath>ensuremath<pensuremath>FUNDING: Doctoral Studentship for NT (Institute for Life Sciences (University of Southampton) & University Hospital Southampton), UKRI TAS Pump-Priming Grant (TAS_PP_00167).ensuremath</pensuremath>
Ansari, Aamir Ahmad; Ramchurn, Gopal; Nguyen, Tan Viet Tuyen
Beyond text: multi-modal LLM in human robot interaction Proceedings Article
In: UK AI Research Symposium: A Festival of Ideas (08/09/25 - 09/09/25), 2025.
@inproceedings{soton505659,
title = {Beyond text: multi-modal LLM in human robot interaction},
author = {Aamir Ahmad Ansari and Gopal Ramchurn and Tan Viet Tuyen Nguyen},
url = {https://eprints.soton.ac.uk/505659/},
year = {2025},
date = {2025-09-01},
booktitle = {UK AI Research Symposium: A Festival of Ideas (08/09/25 - 09/09/25)},
abstract = {Multimodal interaction plays a vital role in Human-Robot Interaction (HRI), enabling robots to communicate with humans through multiple channels. This study introduces a novel approach to enhance such interactions by treating images and human motion as distinct foreign languages, in addition to text. In the proposed framework, vector quantization is employed to convert multimodal inputs such as images and human motions to an aligned set of tokens. A Large Language Model (LLM) is then pre-trained with the use of Low-Rank Adaptation (LoRA) and instruction-tuned on a dialogue dataset that incorporates both image and motion context. The proposed multimodal LLM framework aims to equip robots with the ability to understand and respond to complex human queries through multimodal inputs and outputs, enabling more natural and effective interactions.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Multimodal interaction plays a vital role in Human-Robot Interaction (HRI), enabling robots to communicate with humans through multiple channels. This study introduces a novel approach to enhance such interactions by treating images and human motion as distinct foreign languages, in addition to text. In the proposed framework, vector quantization is employed to convert multimodal inputs such as images and human motions to an aligned set of tokens. A Large Language Model (LLM) is then pre-trained with the use of Low-Rank Adaptation (LoRA) and instruction-tuned on a dialogue dataset that incorporates both image and motion context. The proposed multimodal LLM framework aims to equip robots with the ability to understand and respond to complex human queries through multimodal inputs and outputs, enabling more natural and effective interactions.
Le, Thien Doanh; Nguyen, Tan Viet Tuyen; Duy, Tan Le; Ramchurn, Gopal
A multimodal large language model framework for gesture generation in social robots Proceedings Article
In: BEAR Workshop, The 2025 IEEE International Conference on Robot and Human Interactive Communication (25/08/25 - 29/08/25), 2025.
@inproceedings{soton506590,
title = {A multimodal large language model framework for gesture generation in social robots},
author = {Thien Doanh Le and Tan Viet Tuyen Nguyen and Tan Le Duy and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/506590/},
year = {2025},
date = {2025-08-01},
booktitle = {BEAR Workshop, The 2025 IEEE International Conference on Robot and Human Interactive Communication (25/08/25 - 29/08/25)},
abstract = {Non-verbal gestures play a crucial role in social robots, enabling them to signal their intentions to users during human?robot interaction (HRI). While recent research in this domain has primarily focused on robot gesture generation, there remains a limited number of studies on multimodal generation frameworks, where generated gestures are harmonized with other generated modalities, to better convey the robot?s intention to users via a wider range of communication channels. Inspired by recent advancements in multimodal large language models (MLLMs), we propose a novel framework that integrates motion generation models with existing MLLMs to produce high-quality 3D motions without the need for extensive multimodal training. Our framework comprises three key components: a Denoising Diffusion Motion Generation (DDMG) module that maps text descriptions to motion sequences using a diffusion-based approach; a Motion Decoding Alignment (MDA) module that refines motion representations by incorporating signal embeddings generated by an LLM; and a Fusion Module (FM) that integrates motion features trained from previous phases to enhance coherence and realism. We conducted a series of experiments on a publicly available dataset to evaluate the efficiency of the proposed framework in terms of motion quality, diversity, and semantic alignment. The results suggest that our multimodal approach can serve as a powerful controller for robot gesture generation, offering a more scalable and effective solution, particularly for social HRI.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Non-verbal gestures play a crucial role in social robots, enabling them to signal their intentions to users during human?robot interaction (HRI). While recent research in this domain has primarily focused on robot gesture generation, there remains a limited number of studies on multimodal generation frameworks, where generated gestures are harmonized with other generated modalities, to better convey the robot?s intention to users via a wider range of communication channels. Inspired by recent advancements in multimodal large language models (MLLMs), we propose a novel framework that integrates motion generation models with existing MLLMs to produce high-quality 3D motions without the need for extensive multimodal training. Our framework comprises three key components: a Denoising Diffusion Motion Generation (DDMG) module that maps text descriptions to motion sequences using a diffusion-based approach; a Motion Decoding Alignment (MDA) module that refines motion representations by incorporating signal embeddings generated by an LLM; and a Fusion Module (FM) that integrates motion features trained from previous phases to enhance coherence and realism. We conducted a series of experiments on a publicly available dataset to evaluate the efficiency of the proposed framework in terms of motion quality, diversity, and semantic alignment. The results suggest that our multimodal approach can serve as a powerful controller for robot gesture generation, offering a more scalable and effective solution, particularly for social HRI.
Ramchurn, Gopal; Jones, Matt; Owen, Aled Lloyd; Ehsan, Shoaib; Kiden, Sarah; Eke, Damian; McStay, Andrew; Reitmaier, Thomas; Raju, Dani Kalarikalayil; Castenada, Nestor; Sailaja, Neelima; Faith, Becky; Hermann, Antony; Kalra, Kanika; Hawes, Ben; Adams, Rachel; Meier, Patrick; Sharma, Gaurav; Vishwarupe, Varad
Responsible AI. To enable flourishing in and by Low- and Middle- Income Countries (LMICs) Book Section
In: Responsible Ai UK, University of Southampton, 2025.
@incollection{soton506045,
title = {Responsible AI. To enable flourishing in and by Low- and Middle- Income Countries (LMICs)},
author = {Gopal Ramchurn and Matt Jones and Aled Lloyd Owen and Shoaib Ehsan and Sarah Kiden and Damian Eke and Andrew McStay and Thomas Reitmaier and Dani Kalarikalayil Raju and Nestor Castenada and Neelima Sailaja and Becky Faith and Antony Hermann and Kanika Kalra and Ben Hawes and Rachel Adams and Patrick Meier and Gaurav Sharma and Varad Vishwarupe},
url = {https://eprints.soton.ac.uk/506045/},
year = {2025},
date = {2025-07-01},
booktitle = {Responsible Ai UK},
publisher = {University of Southampton},
abstract = {This report explores how governments and industry in Low- and Middle-Income Countries (LMICs) might adopt AI to serve the values, needs and aspirations of individuals and communities, and drive the achievement of development goals. It draws on RAi UK projects working with partners in India and Indonesia and in countries in Africa and in South America, and identifies challenges, early solutions and policy implications. The aim is to identify priority areas and questions for RAi UK and other researchers and organisations to explore further. In this report we focus on the emerging challenges and other outputs of research on AI for LMICs. We convened leaders and members of projects that explore embedding AI tools and resources that are sensitive to local, cultural norms and end-user experience through co-creation with communities.},
keywords = {},
pubstate = {published},
tppubtype = {incollection}
}
This report explores how governments and industry in Low- and Middle-Income Countries (LMICs) might adopt AI to serve the values, needs and aspirations of individuals and communities, and drive the achievement of development goals. It draws on RAi UK projects working with partners in India and Indonesia and in countries in Africa and in South America, and identifies challenges, early solutions and policy implications. The aim is to identify priority areas and questions for RAi UK and other researchers and organisations to explore further. In this report we focus on the emerging challenges and other outputs of research on AI for LMICs. We convened leaders and members of projects that explore embedding AI tools and resources that are sensitive to local, cultural norms and end-user experience through co-creation with communities.
Abioye, Ayodeji O.; Hunt, William; Gu, Yue; Schneiders, Eike; Naiseh, Mohammad; Archibald, Blair; Sevegnani, Michele; Ramchurn, Sarvapali D.; Fischer, Joel E.; Soorati, Mohammad D.
A user study evaluation of predictive formal modelling at runtime in human-swarm interaction Journal Article
In: ACM Transactions on Human-Robot Interaction, vol. 14, no. 4, 2025.
@article{soton501672,
title = {A user study evaluation of predictive formal modelling at runtime in human-swarm interaction},
author = {Ayodeji O. Abioye and William Hunt and Yue Gu and Eike Schneiders and Mohammad Naiseh and Blair Archibald and Michele Sevegnani and Sarvapali D. Ramchurn and Joel E. Fischer and Mohammad D. Soorati},
url = {https://eprints.soton.ac.uk/501672/},
year = {2025},
date = {2025-06-01},
journal = {ACM Transactions on Human-Robot Interaction},
volume = {14},
number = {4},
abstract = {Formal modelling is often used as part of the design and testing process of software development to ensure that components operate within suitable bounds even in unexpected circumstances. We conducted a user study evaluation of predictive formal modelling (PFM) at runtime in a human-swarm mission to determine the benefit of predictive formal modelling on performance and human-swarm interaction. 180 participants were recruited to perform the role of aerial swarm operators delivering parcels to target locations in a simulation environment. The PFM model was integrated into the simulation software to inform the operator of the estimated mission completion time given the current number of drones deployed. The operator could increase the number of parcels delivered in any time step by adding drones, which also increased costs, thus requiring the use of the minimum number of drones necessary to complete the task in the given time. We collected user feedback using standard survey questionnaires and measured performance using data obtained from the Human And Robot Interactive Swarm (HARIS) simulator. Our results show that PFM increased the performance of the human swarm team without significantly increasing the operators? workload or affecting the system's usability.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Formal modelling is often used as part of the design and testing process of software development to ensure that components operate within suitable bounds even in unexpected circumstances. We conducted a user study evaluation of predictive formal modelling (PFM) at runtime in a human-swarm mission to determine the benefit of predictive formal modelling on performance and human-swarm interaction. 180 participants were recruited to perform the role of aerial swarm operators delivering parcels to target locations in a simulation environment. The PFM model was integrated into the simulation software to inform the operator of the estimated mission completion time given the current number of drones deployed. The operator could increase the number of parcels delivered in any time step by adding drones, which also increased costs, thus requiring the use of the minimum number of drones necessary to complete the task in the given time. We collected user feedback using standard survey questionnaires and measured performance using data obtained from the Human And Robot Interactive Swarm (HARIS) simulator. Our results show that PFM increased the performance of the human swarm team without significantly increasing the operators? workload or affecting the system's usability.
Deshmukh, Jayati; Liang, Zijie; Yazdanpanah, Vahid; Stein, Sebastian; Ramchurn, Savapali D.
Serious games for ethical preference elicitation Proceedings Article
In: AAMAS '25: Proceedings of the 24th International Conference on Autonomous Agents and Multiagent Systems, pp. 3018 – 3020, Association for Computing Machinery, 2025.
@inproceedings{soton498743b,
title = {Serious games for ethical preference elicitation},
author = {Jayati Deshmukh and Zijie Liang and Vahid Yazdanpanah and Sebastian Stein and Savapali D. Ramchurn},
url = {https://eprints.soton.ac.uk/498743/},
year = {2025},
date = {2025-06-01},
booktitle = {AAMAS '25: Proceedings of the 24th International Conference on Autonomous Agents and Multiagent Systems},
pages = {3018 – 3020},
publisher = {Association for Computing Machinery},
abstract = {Autonomous agents acting on behalf of humans must act according to their ethical preferences. However, ethical preferences are latent and abstract and thus it is challenging to elicit them. To address this, we present a serious game that helps elicit ethical preferences in a more dynamic and engaging way than traditional methods such as questionnaires or simple dilemmas.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Autonomous agents acting on behalf of humans must act according to their ethical preferences. However, ethical preferences are latent and abstract and thus it is challenging to elicit them. To address this, we present a serious game that helps elicit ethical preferences in a more dynamic and engaging way than traditional methods such as questionnaires or simple dilemmas.
Deshmukh, Jayati; Liang, Zijie; Yazdanpanah, Vahid; Stein, Sebastian; Ramchurn, Sarvalpali D.
Serious games for ethical preference elicitation Proceedings Article
In: AAMAS - 2025 : The 24th International Conference on Autonomous Agents and Multiagent Systems (19/05/25 - 23/05/25), 2025.
@inproceedings{soton498743,
title = {Serious games for ethical preference elicitation},
author = {Jayati Deshmukh and Zijie Liang and Vahid Yazdanpanah and Sebastian Stein and Sarvalpali D. Ramchurn},
url = {https://eprints.soton.ac.uk/498743/},
year = {2025},
date = {2025-05-01},
booktitle = {AAMAS - 2025 : The 24th International Conference on Autonomous Agents and Multiagent Systems (19/05/25 - 23/05/25)},
abstract = {Autonomous agents acting on behalf of humans must act according to their ethical preferences. However, ethical preferences are latent and abstract and thus it is challenging to elicit them. To address this, we present a serious game that helps elicit ethical preferences in a more dynamic and engaging way than traditional methods such as questionnaires or simple dilemmas.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Autonomous agents acting on behalf of humans must act according to their ethical preferences. However, ethical preferences are latent and abstract and thus it is challenging to elicit them. To address this, we present a serious game that helps elicit ethical preferences in a more dynamic and engaging way than traditional methods such as questionnaires or simple dilemmas.
Abioye, Ayodeji; Hunt, William; Schneiders, Eike; Gu, Yue; Naiseh, Mohammad; Archibald, Blair; Sevegnani, Michele; Ramchurn, Gopal; Fischer, Joel; Soorati, Mohammad
Research Data for Paper: A User Study Evaluation of Predictive Formal Modelling at Runtime in Human-Swarm Interaction Miscellaneous
2025.
@misc{soton500788,
title = {Research Data for Paper: A User Study Evaluation of Predictive Formal Modelling at Runtime in Human-Swarm Interaction},
author = {Ayodeji Abioye and William Hunt and Eike Schneiders and Yue Gu and Mohammad Naiseh and Blair Archibald and Michele Sevegnani and Gopal Ramchurn and Joel Fischer and Mohammad Soorati},
url = {https://eprints.soton.ac.uk/500788/},
year = {2025},
date = {2025-03-01},
publisher = {figshare},
abstract = {ensuremath<span style='white-space:pre-line'ensuremath>This dataset is for the user study conducted to evaluate the performance benefits of predictive formal modelling (PFM) at runtime in a human-swarm interaction experiment. We recruited 180 participants to perform the role of aerial swarm operators conducting a drone delivery mission in a simulation environment using the Human And Robot Interactive Swarm (HARIS) simulator.ensuremath</spanensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
ensuremath<span style='white-space:pre-line'ensuremath>This dataset is for the user study conducted to evaluate the performance benefits of predictive formal modelling (PFM) at runtime in a human-swarm interaction experiment. We recruited 180 participants to perform the role of aerial swarm operators conducting a drone delivery mission in a simulation environment using the Human And Robot Interactive Swarm (HARIS) simulator.ensuremath</spanensuremath>
Ismagilov, Timur; Ferrarini, Bruno; Milford, Michael; Tuyen, Nguyen Tan Viet; Ramchurn, Sarvapali D.; Ehsan, Shoaib
On motion blur and deblurring in visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 10, no. 5, pp. 4746–4753, 2025.
@article{soton507348,
title = {On motion blur and deblurring in visual place recognition},
author = {Timur Ismagilov and Bruno Ferrarini and Michael Milford and Nguyen Tan Viet Tuyen and Sarvapali D. Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/507348/},
year = {2025},
date = {2025-03-01},
journal = {IEEE Robotics and Automation Letters},
volume = {10},
number = {5},
pages = {4746–4753},
abstract = {ensuremath<pensuremath>Visual Place Recognition (VPR) in mobile robotics enables robots to localize themselves by recognizing previously visited locations using visual data. While the reliability of VPR methods has been extensively studied under conditions such as changes in illumination, season, weather and viewpoint, the impact of motion blur is relatively unexplored despite its relevance not only in rapid motion scenarios but also in low-light conditions where longer exposure times are necessary. Similarly, the role of image deblurring in enhancing VPR performance under motion blur has received limited attention so far. This letter bridges these gaps by introducing a new benchmark designed to evaluate VPR performance under the influence of motion blur and image deblurring. The benchmark includes three datasets that encompass a wide range of motion blur intensities, providing a comprehensive platform for analysis. Experimental results with several well-established VPR and image deblurring methods provide new insights into the effects of motion blur and the potential improvements achieved through deblurring. Building on these findings, the letter proposes adaptive deblurring strategies for VPR, designed to effectively manage motion blur in dynamic, real-world scenarios.ensuremath</pensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Visual Place Recognition (VPR) in mobile robotics enables robots to localize themselves by recognizing previously visited locations using visual data. While the reliability of VPR methods has been extensively studied under conditions such as changes in illumination, season, weather and viewpoint, the impact of motion blur is relatively unexplored despite its relevance not only in rapid motion scenarios but also in low-light conditions where longer exposure times are necessary. Similarly, the role of image deblurring in enhancing VPR performance under motion blur has received limited attention so far. This letter bridges these gaps by introducing a new benchmark designed to evaluate VPR performance under the influence of motion blur and image deblurring. The benchmark includes three datasets that encompass a wide range of motion blur intensities, providing a comprehensive platform for analysis. Experimental results with several well-established VPR and image deblurring methods provide new insights into the effects of motion blur and the potential improvements achieved through deblurring. Building on these findings, the letter proposes adaptive deblurring strategies for VPR, designed to effectively manage motion blur in dynamic, real-world scenarios.ensuremath</pensuremath>
Grainge, Oliver Edward; Milford, Michael J.; Bodala, Indu; Ramchurn, Sarvapali D.; Ehsan, Shoaib
Structured pruning for efficient visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 10, no. 2, pp. 2024–2031, 2025.
@article{soton502843,
title = {Structured pruning for efficient visual place recognition},
author = {Oliver Edward Grainge and Michael J. Milford and Indu Bodala and Sarvapali D. Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/502843/},
year = {2025},
date = {2025-02-01},
journal = {IEEE Robotics and Automation Letters},
volume = {10},
number = {2},
pages = {2024–2031},
abstract = {Visual Place Recognition (VPR) is fundamental for the global re-localization of robots and devices, enabling them to recognize previously visited locations based on visual inputs. This capability is crucial for maintaining accurate mapping and localization over large areas. Given that VPR methods need to operate in real-time on embedded systems, it is critical to optimize these systems for minimal resource consumption. While the most efficient VPR approaches employ standard convolutional backbones with fixed descriptor dimensions, these often lead to redundancy in the embedding space as well as in the network architecture. Our work introduces a novel structured pruning method, to not only streamline common VPR architectures but also to strategically remove redundancies within the feature embedding space. This dual focus significantly enhances the efficiency of the system, reducing both map and model memory requirements and decreasing feature extraction and retrieval latencies. Our approach has reduced memory usage and latency by 21% and 16%, respectively, across models, while minimally impacting recall@1 accuracy by less than 1%. This significant improvement enhances real-time applications on edge devices with negligible accuracy loss.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Visual Place Recognition (VPR) is fundamental for the global re-localization of robots and devices, enabling them to recognize previously visited locations based on visual inputs. This capability is crucial for maintaining accurate mapping and localization over large areas. Given that VPR methods need to operate in real-time on embedded systems, it is critical to optimize these systems for minimal resource consumption. While the most efficient VPR approaches employ standard convolutional backbones with fixed descriptor dimensions, these often lead to redundancy in the embedding space as well as in the network architecture. Our work introduces a novel structured pruning method, to not only streamline common VPR architectures but also to strategically remove redundancies within the feature embedding space. This dual focus significantly enhances the efficiency of the system, reducing both map and model memory requirements and decreasing feature extraction and retrieval latencies. Our approach has reduced memory usage and latency by 21% and 16%, respectively, across models, while minimally impacting recall@1 accuracy by less than 1%. This significant improvement enhances real-time applications on edge devices with negligible accuracy loss.
Kelly, Thomas Graham
Self-organised communication-aware control structures for robot swarms PhD Thesis
University of Southampton, 2025.
@phdthesis{soton502200,
title = {Self-organised communication-aware control structures for robot swarms},
author = {Thomas Graham Kelly},
url = {https://eprints.soton.ac.uk/502200/},
year = {2025},
date = {2025-01-01},
publisher = {University of Southampton},
school = {University of Southampton},
abstract = {Robotic swarms are complex systems that rely on local communication between individual members of the swarm to spread information about their state and the environment. This information informs decisions and aids in tasks such as exploration and mapping. When communications break down between members of a swarm, it can become difficult to maintain accurate and up-to-date information about the state of the swarm and the environment. This problem is pertinent when humans are involved and may act as operators or teammates of the swarm. Here it is vital that the swarm can coordinate to distil and disseminate the vast amounts of information collected to the humans and throughout the swarm effectively, to maintain situational awareness.ensuremath<br/ensuremath>ensuremath<br/ensuremath>The collective decision-making of a swarm is one aspect that relies heavily on the ability to share information and observations to reach a swarm-wide consensus. This thesis investigates how communication constraints affect the swarm?s ability to reach a consensus and implement a communication-aware coordination strategy to mitigate these effects. We propose the communication-constrained collective decision-making problem and compare the performance of several collective decision-making strategies, enhanced with our coordination algorithm. We find that using such an approach improves the speed of a swarm to reach a consensus.ensuremath<br/ensuremath>ensuremath<br/ensuremath>Following on from this work, we examine a hybrid swarm system in a communication-limited environment. While swarms are traditionally considered decentralised systems, recent approaches have integrated decentralised and centralised control into a swarm system. We study the trade-offs in performance and communication in a hybrid system that can vary its control structure. We find that a higher level of centralisation does not guarantee higher performance and study how communication with a human operator is affected by the control structure. This work is extended to assess the feasibility of enabling a swarm system to learn the optimal control structure on the fly, according to mission requirements.},
keywords = {},
pubstate = {published},
tppubtype = {phdthesis}
}
Robotic swarms are complex systems that rely on local communication between individual members of the swarm to spread information about their state and the environment. This information informs decisions and aids in tasks such as exploration and mapping. When communications break down between members of a swarm, it can become difficult to maintain accurate and up-to-date information about the state of the swarm and the environment. This problem is pertinent when humans are involved and may act as operators or teammates of the swarm. Here it is vital that the swarm can coordinate to distil and disseminate the vast amounts of information collected to the humans and throughout the swarm effectively, to maintain situational awareness.ensuremath<br/ensuremath>ensuremath<br/ensuremath>The collective decision-making of a swarm is one aspect that relies heavily on the ability to share information and observations to reach a swarm-wide consensus. This thesis investigates how communication constraints affect the swarm?s ability to reach a consensus and implement a communication-aware coordination strategy to mitigate these effects. We propose the communication-constrained collective decision-making problem and compare the performance of several collective decision-making strategies, enhanced with our coordination algorithm. We find that using such an approach improves the speed of a swarm to reach a consensus.ensuremath<br/ensuremath>ensuremath<br/ensuremath>Following on from this work, we examine a hybrid swarm system in a communication-limited environment. While swarms are traditionally considered decentralised systems, recent approaches have integrated decentralised and centralised control into a swarm system. We study the trade-offs in performance and communication in a hybrid system that can vary its control structure. We find that a higher level of centralisation does not guarantee higher performance and study how communication with a human operator is affected by the control structure. This work is extended to assess the feasibility of enabling a swarm system to learn the optimal control structure on the fly, according to mission requirements.
Kiden, Sarah; Parisio, Isabela; Neff, Gina; Ramchurn, Gopal
AI management essentials (AIME) consultation response Technical Report
no. 10.5258/SOTON/PP0116, 2025.
@techreport{soton501075,
title = {AI management essentials (AIME) consultation response},
author = {Sarah Kiden and Isabela Parisio and Gina Neff and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/501075/},
year = {2025},
date = {2025-01-01},
number = {10.5258/SOTON/PP0116},
publisher = {University of Southampton},
abstract = {We are submitting this response to the Department for Science, Innovation and Technology (DSIT) consultation on the new AI Management Essentials (AIME) tool1 on behalf of Responsible AI UK (RAi UK), an open and multidisciplinary network that brings together researchers from across the four nations of the UK to understand how we should shape the development of AI to benefit people, communities and society. To arrive at this response, we sent out a call to Principal Investigators and Co-Investigators of RAi UK funded projects2 to contribute to this consultation. What follows is a synthesis of the responses we received from our research community. Overall, RAi UK sees the AIME tool as a valuable first step for Small to Medium Sized Enterprises (SMEs) and Startups to adopt towards implementing robust and responsible AI governance practices. Our contributions below aim to improve the tool?s usability and impact.},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
We are submitting this response to the Department for Science, Innovation and Technology (DSIT) consultation on the new AI Management Essentials (AIME) tool1 on behalf of Responsible AI UK (RAi UK), an open and multidisciplinary network that brings together researchers from across the four nations of the UK to understand how we should shape the development of AI to benefit people, communities and society. To arrive at this response, we sent out a call to Principal Investigators and Co-Investigators of RAi UK funded projects2 to contribute to this consultation. What follows is a synthesis of the responses we received from our research community. Overall, RAi UK sees the AIME tool as a valuable first step for Small to Medium Sized Enterprises (SMEs) and Startups to adopt towards implementing robust and responsible AI governance practices. Our contributions below aim to improve the tool?s usability and impact.
Kiden, Sarah; Yazdanpanah, Vahid; Mestre, Rafael; Iusmen, Ingi; Atkinson, Joe; Parisio, Isabela; Ramchurn, Gopal; Stein, Sebastian
Human Rights and the Regulation of AI Written Evidence Technical Report
no. 10.5258/SOTON/PP0152, 2025.
@techreport{soton506667,
title = {Human Rights and the Regulation of AI Written Evidence},
author = {Sarah Kiden and Vahid Yazdanpanah and Rafael Mestre and Ingi Iusmen and Joe Atkinson and Isabela Parisio and Gopal Ramchurn and Sebastian Stein},
url = {https://eprints.soton.ac.uk/506667/},
year = {2025},
date = {2025-01-01},
number = {10.5258/SOTON/PP0152},
publisher = {University of Southampton},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
2024
Thavanesan, Navamayooran; Farahi, Arya; Parfitt, Charlotte; Belkhatir, Zehor; Azim, Tayyaba; Vallejos, Elvira Perez; Walters, Zoë; Ramchurn, Sarvapali; Underwood, Timothy J.; Vigneswaran, Ganesh
Insights from explainable AI in oesophageal cancer team decisions Journal Article
In: Computers in Biology and Medicine, vol. 180, 2024, (For the purpose of open access, the authors have applied a Creative Commons attribution license (CC-BY) to any Author Accepted Manuscript version arising from this submission.).
@article{soton493238,
title = {Insights from explainable AI in oesophageal cancer team decisions},
author = {Navamayooran Thavanesan and Arya Farahi and Charlotte Parfitt and Zehor Belkhatir and Tayyaba Azim and Elvira Perez Vallejos and Zoë Walters and Sarvapali Ramchurn and Timothy J. Underwood and Ganesh Vigneswaran},
url = {https://eprints.soton.ac.uk/493238/},
year = {2024},
date = {2024-08-01},
journal = {Computers in Biology and Medicine},
volume = {180},
abstract = {ensuremath<pensuremath>Background: clinician-led quality control into oncological decision-making is crucial for optimising patient care. Explainable artificial intelligence (XAI) techniques provide data-driven approaches to unravel how clinical variables influence this decision-making. We applied global XAI techniques to examine the impact of key clinical decision-drivers when mapped by a machine learning (ML) model, on the likelihood of receiving different oesophageal cancer (OC) treatment modalities by the multidisciplinary team (MDT).ensuremath</pensuremath>ensuremath<pensuremath>Methods: retrospective analysis of 893 OC patients managed between 2010 and 2022 at our tertiary unit, used a random forests (RF) classifier to predict four possible treatment pathways as determined by the MDT: neoadjuvant chemotherapy followed by surgery (NACT + S), neoadjuvant chemoradiotherapy followed by surgery (NACRT + S), surgery-alone, and palliative management. Variable importance and partial dependence (PD) analyses then examined the influence of targeted high-ranking clinical variables within the ML model on treatment decisions as a surrogate model of the MDT decision-making dynamic.�ensuremath</pensuremath>ensuremath<pensuremath>Results: amongst guideline-variables known to determine treatments, such as Tumour-Node-Metastasis (TNM) staging, age also proved highly important to the RF model (16.1 % of total importance) on variable importance analysis. PD subsequently revealed that predicted probabilities for all treatment modalities change significantly after 75 years (p < 0.001). Likelihood of surgery-alone and palliative therapies increased for patients aged 75?85yrs but lowered for NACT/NACRT. Performance status divided patients into two clusters which influenced all predicted outcomes in conjunction with age.�ensuremath</pensuremath>ensuremath<pensuremath>Conclusion: XAI techniques delineate the relationship between clinical factors and OC treatment decisions. These techniques identify advanced age as heavily influencing decisions based on our model with a greater role in patients with specific tumour characteristics. This study methodology provides the means for exploring conscious/subconscious bias and interrogating inconsistencies in team-based decision-making within the era of AI-driven decision support.ensuremath</pensuremath>},
note = {For the purpose of open access, the authors have applied a Creative Commons attribution license (CC-BY) to any Author Accepted Manuscript version arising from this submission.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Background: clinician-led quality control into oncological decision-making is crucial for optimising patient care. Explainable artificial intelligence (XAI) techniques provide data-driven approaches to unravel how clinical variables influence this decision-making. We applied global XAI techniques to examine the impact of key clinical decision-drivers when mapped by a machine learning (ML) model, on the likelihood of receiving different oesophageal cancer (OC) treatment modalities by the multidisciplinary team (MDT).ensuremath</pensuremath>ensuremath<pensuremath>Methods: retrospective analysis of 893 OC patients managed between 2010 and 2022 at our tertiary unit, used a random forests (RF) classifier to predict four possible treatment pathways as determined by the MDT: neoadjuvant chemotherapy followed by surgery (NACT + S), neoadjuvant chemoradiotherapy followed by surgery (NACRT + S), surgery-alone, and palliative management. Variable importance and partial dependence (PD) analyses then examined the influence of targeted high-ranking clinical variables within the ML model on treatment decisions as a surrogate model of the MDT decision-making dynamic.�ensuremath</pensuremath>ensuremath<pensuremath>Results: amongst guideline-variables known to determine treatments, such as Tumour-Node-Metastasis (TNM) staging, age also proved highly important to the RF model (16.1 % of total importance) on variable importance analysis. PD subsequently revealed that predicted probabilities for all treatment modalities change significantly after 75 years (p < 0.001). Likelihood of surgery-alone and palliative therapies increased for patients aged 75?85yrs but lowered for NACT/NACRT. Performance status divided patients into two clusters which influenced all predicted outcomes in conjunction with age.�ensuremath</pensuremath>ensuremath<pensuremath>Conclusion: XAI techniques delineate the relationship between clinical factors and OC treatment decisions. These techniques identify advanced age as heavily influencing decisions based on our model with a greater role in patients with specific tumour characteristics. This study methodology provides the means for exploring conscious/subconscious bias and interrogating inconsistencies in team-based decision-making within the era of AI-driven decision support.ensuremath</pensuremath>
Naiseh, Mohammad; Webb, Catherine; Underwood, Tim; Ramchurn, Gopal; Walters, Zoe; Thavanesan, Navamayooran; Vigneswaran, Ganesh
XAI for group-AI interaction: towards collaborative and inclusive explanations Proceedings Article
In: Longo, Luca; Liu, Weiru; Montavon, Gregoire (Ed.): Joint Proceedings of the xAI 2024 Late-breaking Work, Demos and Doctoral Consortium co-located with the 2nd World Conference on eXplainable Artificial Intelligence (xAI 2024), pp. 249–256, CEUR Workshop Proceedings, 2024.
@inproceedings{soton497829,
title = {XAI for group-AI interaction: towards collaborative and inclusive explanations},
author = {Mohammad Naiseh and Catherine Webb and Tim Underwood and Gopal Ramchurn and Zoe Walters and Navamayooran Thavanesan and Ganesh Vigneswaran},
editor = {Luca Longo and Weiru Liu and Gregoire Montavon},
url = {https://eprints.soton.ac.uk/497829/},
year = {2024},
date = {2024-07-01},
booktitle = {Joint Proceedings of the xAI 2024 Late-breaking Work, Demos and Doctoral Consortium co-located with the 2nd World Conference on eXplainable Artificial Intelligence (xAI 2024)},
volume = {3793},
pages = {249–256},
publisher = {CEUR Workshop Proceedings},
abstract = {ensuremath<pensuremath>The increasing integration of Machine Learning (ML) into decision-making across various sectors has raised concerns about ethics, legality, explainability, and safety, highlighting the necessity of human oversight. In response, eXplainable AI (XAI) has emerged as a means to enhance transparency by providing insights into ML model decisions and offering humans an understanding of the underlying logic. Despite its potential, existing XAI models often lack practical usability and fail to improve human-AI performance, as they may introduce issues such as overreliance. This underscores the need for further research in Human-Centered XAI to improve the usability of current XAI methods. Notably, much of the current research focuses on one-to-one interactions between the XAI and individual decision-makers, overlooking the dynamics of many-to-one relationships in real-world scenarios where groups of humans collaborate using XAI in collective decision-making. In this late-breaking work, we draw upon current work in Human-Centered XAI research and discuss how XAI design could be transitioned to group-AI interaction. We discuss four potential challenges in the transition of XAI from human-AI interaction to group-AI interaction. This paper contributes to advancing the field of Human-Centered XAI and facilitates the discussion on group-XAI interaction, calling for further research in this area.ensuremath</pensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
ensuremath<pensuremath>The increasing integration of Machine Learning (ML) into decision-making across various sectors has raised concerns about ethics, legality, explainability, and safety, highlighting the necessity of human oversight. In response, eXplainable AI (XAI) has emerged as a means to enhance transparency by providing insights into ML model decisions and offering humans an understanding of the underlying logic. Despite its potential, existing XAI models often lack practical usability and fail to improve human-AI performance, as they may introduce issues such as overreliance. This underscores the need for further research in Human-Centered XAI to improve the usability of current XAI methods. Notably, much of the current research focuses on one-to-one interactions between the XAI and individual decision-makers, overlooking the dynamics of many-to-one relationships in real-world scenarios where groups of humans collaborate using XAI in collective decision-making. In this late-breaking work, we draw upon current work in Human-Centered XAI research and discuss how XAI design could be transitioned to group-AI interaction. We discuss four potential challenges in the transition of XAI from human-AI interaction to group-AI interaction. This paper contributes to advancing the field of Human-Centered XAI and facilitates the discussion on group-XAI interaction, calling for further research in this area.ensuremath</pensuremath>
Early, Joseph Arthur
Interpretable multiple instance learning PhD Thesis
University of Southampton, 2024.
@phdthesis{soton490767,
title = {Interpretable multiple instance learning},
author = {Joseph Arthur Early},
url = {https://eprints.soton.ac.uk/490767/},
year = {2024},
date = {2024-06-01},
publisher = {University of Southampton},
school = {University of Southampton},
abstract = {With the rising use of Artificial Intelligence (AI) and Machine Learning (ML) methods, there comes an increasing need to understand how automated systems make decisions. Interpretable ML provides insight into the underlying reasoning behind AI and ML models while not stifling their predictive performance. Doing so is important for many reasons, such as facilitating trust, increasing transparency, and providing improved collaboration and control through a better understanding of automated decision-making. Interpretability is very relevant across many ML paradigms and application domains. Multiple Instance Learning (MIL) is an ML paradigm where data are grouped into bags of instances, and only the bags are labelled (rather than each instance). This is beneficial in alleviating expensive labelling procedures and can be used to exploit the underlying structure of data. This thesis investigates how interpretability can be achieved within MIL. It begins with a formalisation of interpretable MIL, and then proposes a suite of model-agnostic post-hoc methods. This work is then extended to the specific application domain of high-resolution satellite imagery, using novel inherently interpretable MIL approaches that operate at multiple resolutions. Following on from work in the vision domain, new methods for interpretable MIL are developed for sequential data. First, it is explored in the domain of Reward Modelling (RM) for Reinforcement Learning (RL), demonstrating that interpretable MIL can be used to not only understand a model but also improve its predictive performance. This is mirrored in the application of interpretable MIL to Time Series Classification (TSC), where it is integrated into state-of-the-art methods and is able to improve both their interpretability and predictive performance. The integration into existing models to provide inherent interpretability means these benefits are delivered with little additional computational cost. ensuremath<br/ensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {phdthesis}
}
With the rising use of Artificial Intelligence (AI) and Machine Learning (ML) methods, there comes an increasing need to understand how automated systems make decisions. Interpretable ML provides insight into the underlying reasoning behind AI and ML models while not stifling their predictive performance. Doing so is important for many reasons, such as facilitating trust, increasing transparency, and providing improved collaboration and control through a better understanding of automated decision-making. Interpretability is very relevant across many ML paradigms and application domains. Multiple Instance Learning (MIL) is an ML paradigm where data are grouped into bags of instances, and only the bags are labelled (rather than each instance). This is beneficial in alleviating expensive labelling procedures and can be used to exploit the underlying structure of data. This thesis investigates how interpretability can be achieved within MIL. It begins with a formalisation of interpretable MIL, and then proposes a suite of model-agnostic post-hoc methods. This work is then extended to the specific application domain of high-resolution satellite imagery, using novel inherently interpretable MIL approaches that operate at multiple resolutions. Following on from work in the vision domain, new methods for interpretable MIL are developed for sequential data. First, it is explored in the domain of Reward Modelling (RM) for Reinforcement Learning (RL), demonstrating that interpretable MIL can be used to not only understand a model but also improve its predictive performance. This is mirrored in the application of interpretable MIL to Time Series Classification (TSC), where it is integrated into state-of-the-art methods and is able to improve both their interpretability and predictive performance. The integration into existing models to provide inherent interpretability means these benefits are delivered with little additional computational cost. ensuremath<br/ensuremath>
Kiden, Sarah; Stahl, Bernd; Townsend, Beverley; Maple, Carsten; Vincent, Charles; Sampson, Fraser; Gilbert, Geoff; Smith, Helen; Deshmukh, Jayati; Ross, Jen; Williams, Jennifer; Rincon, Jesus Martinez; Lisinska, Justyna; O?Shea, Karen; Abreu, Márjory Da Costa; Bencomo, Nelly; Deb, Oishi; Winter, Peter; Li, Phoebe; Torr, Philip; Lau, Pin Lean; Iniesta, Raquel; Ramchurn, Gopal; Stein, Sebastian; Yazdanpanah, Vahid
Responsible AI governance: A response to UN interim report on governing AI for humanity Technical Report
no. 10.5258/SOTON/PP0057, 2024.
@techreport{soton488908,
title = {Responsible AI governance: A response to UN interim report on governing AI for humanity},
author = {Sarah Kiden and Bernd Stahl and Beverley Townsend and Carsten Maple and Charles Vincent and Fraser Sampson and Geoff Gilbert and Helen Smith and Jayati Deshmukh and Jen Ross and Jennifer Williams and Jesus Martinez Rincon and Justyna Lisinska and Karen O?Shea and Márjory Da Costa Abreu and Nelly Bencomo and Oishi Deb and Peter Winter and Phoebe Li and Philip Torr and Pin Lean Lau and Raquel Iniesta and Gopal Ramchurn and Sebastian Stein and Vahid Yazdanpanah},
url = {https://eprints.soton.ac.uk/488908/},
year = {2024},
date = {2024-03-01},
number = {10.5258/SOTON/PP0057},
publisher = {Public Policy, University of Southampton},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
Abioye, Ayodeji O.; Hunt, William; Gu, Yue; Schneiders, Eike; Naiseh, Mohammad; Fischer, Joel E.; Ramchurn, Sarvapali D.; Soorati, Mohammad D.; Archibald, Blair; Sevegnani, Michele
The effect of predictive formal modelling at runtime on performance in human-swarm interaction Proceedings Article
In: HRI '24: Companion of the 2024 ACM/IEEE International Conference on Human-Robot Interaction, pp. 172?176, Association for Computing Machinery, 2024, (Publisher Copyright: © 2024 Copyright held by the owner/author(s)).
@inproceedings{soton488273,
title = {The effect of predictive formal modelling at runtime on performance in human-swarm interaction},
author = {Ayodeji O. Abioye and William Hunt and Yue Gu and Eike Schneiders and Mohammad Naiseh and Joel E. Fischer and Sarvapali D. Ramchurn and Mohammad D. Soorati and Blair Archibald and Michele Sevegnani},
url = {https://eprints.soton.ac.uk/488273/},
year = {2024},
date = {2024-03-01},
booktitle = {HRI '24: Companion of the 2024 ACM/IEEE International Conference on Human-Robot Interaction},
pages = {172?176},
publisher = {Association for Computing Machinery},
abstract = {Formal Modelling is often used as part of the design and testing process of software development to ensure that components operate within suitable bounds even in unexpected circumstances. In this paper, we use predictive formal modelling (PFM) at runtime in a human-swarm mission and show that this integration can be used to improve the performance of human-swarm teams. We recruited 60 participants to operate a simulated aerial swarm to deliver parcels to target locations. In the PFM condition, operators were informed of the estimated completion times given the number of drones deployed, whereas, in the No-PFM condition, operators did not have this information. The operators could control the mission by adding or removing drones from the mission and thereby, increasing or decreasing the overall mission cost. The evaluation of human-swarm performance relied on four metrics: the task completion time, the number of agents, the number of completed tasks, and the cost per task. Our results show that PFM modelling at runtime improves mission performance without significantly affecting the operator's workload or the system's usability.},
note = {Publisher Copyright:
© 2024 Copyright held by the owner/author(s)},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Formal Modelling is often used as part of the design and testing process of software development to ensure that components operate within suitable bounds even in unexpected circumstances. In this paper, we use predictive formal modelling (PFM) at runtime in a human-swarm mission and show that this integration can be used to improve the performance of human-swarm teams. We recruited 60 participants to operate a simulated aerial swarm to deliver parcels to target locations. In the PFM condition, operators were informed of the estimated completion times given the number of drones deployed, whereas, in the No-PFM condition, operators did not have this information. The operators could control the mission by adding or removing drones from the mission and thereby, increasing or decreasing the overall mission cost. The evaluation of human-swarm performance relied on four metrics: the task completion time, the number of agents, the number of completed tasks, and the cost per task. Our results show that PFM modelling at runtime improves mission performance without significantly affecting the operator's workload or the system's usability.
Soorati, Mohammad D.; Naiseh, Mohammad; Hunt, William; Parnell, Katie; Clark, Jediah; Ramchurn, Sarvapali D.
Enabling trustworthiness in human-swarm systems through a digital twin Book Section
In: Dasgupta, Prithviraj; Llinas, James; Gillespie, Tony; Fouse, Scott; Lawless, William; Mittu, Ranjeev; Sofge, Donlad (Ed.): Putting AI in the Critical Loop: Assured Trust and Autonomy in Human-Machine Teams, pp. 93–125, Academic Press, 2024, (Publisher Copyright: © 2024 Elsevier Inc. All rights reserved.).
@incollection{soton491769,
title = {Enabling trustworthiness in human-swarm systems through a digital twin},
author = {Mohammad D. Soorati and Mohammad Naiseh and William Hunt and Katie Parnell and Jediah Clark and Sarvapali D. Ramchurn},
editor = {Prithviraj Dasgupta and James Llinas and Tony Gillespie and Scott Fouse and William Lawless and Ranjeev Mittu and Donlad Sofge},
url = {https://eprints.soton.ac.uk/491769/},
year = {2024},
date = {2024-02-01},
booktitle = {Putting AI in the Critical Loop: Assured Trust and Autonomy in Human-Machine Teams},
pages = {93–125},
publisher = {Academic Press},
abstract = {Robot swarms are highly dynamic systems that exhibit fault-tolerant behavior in accomplishing given tasks. Applications of swarm robotics are very limited due to the lack of complex decision-making capability. Real-world applications are only possible if we use human supervision to monitor and control the behavior of the swarm. Ensuring that human operators can trust the swarm system is one of the key challenges in human-swarm systems. This chapter presents a digital twin for trustworthy human-swarm teaming. The first element in designing such a simulation platform is to understand the trust requirements to label a human-swarm system as trustworthy. In order to outline the key trust requirements, we interviewed a group of experienced uncrewed aerial vehicle (UAV) operators and collated their suggestions for building and repairing trusts in single and multiple UAV systems. We then performed a survey to gather swarm experts? points of view on creating a taxonomy for explainability in human-swarm systems. This chapter presents a digital twin platform that implements a disaster management use case and has the capacity to meet the extracted trust and explainability requirements.},
note = {Publisher Copyright:
© 2024 Elsevier Inc. All rights reserved.},
keywords = {},
pubstate = {published},
tppubtype = {incollection}
}
Robot swarms are highly dynamic systems that exhibit fault-tolerant behavior in accomplishing given tasks. Applications of swarm robotics are very limited due to the lack of complex decision-making capability. Real-world applications are only possible if we use human supervision to monitor and control the behavior of the swarm. Ensuring that human operators can trust the swarm system is one of the key challenges in human-swarm systems. This chapter presents a digital twin for trustworthy human-swarm teaming. The first element in designing such a simulation platform is to understand the trust requirements to label a human-swarm system as trustworthy. In order to outline the key trust requirements, we interviewed a group of experienced uncrewed aerial vehicle (UAV) operators and collated their suggestions for building and repairing trusts in single and multiple UAV systems. We then performed a survey to gather swarm experts? points of view on creating a taxonomy for explainability in human-swarm systems. This chapter presents a digital twin platform that implements a disaster management use case and has the capacity to meet the extracted trust and explainability requirements.
Thavanesan, Navamayooran; Parfitt, Charlotte; Bodala, Indu; Walters, Zoë; Ramchurn, Sarvapali; Underwood, Timothy; Vigneswaran, Ganesh
Machine learning models for curative and palliative oesophageal cancer treatment pathway prediction Miscellaneous
2024.
@misc{soton497828,
title = {Machine learning models for curative and palliative oesophageal cancer treatment pathway prediction},
author = {Navamayooran Thavanesan and Charlotte Parfitt and Indu Bodala and Zoë Walters and Sarvapali Ramchurn and Timothy Underwood and Ganesh Vigneswaran},
url = {https://eprints.soton.ac.uk/497828/},
year = {2024},
date = {2024-01-01},
journal = {European Journal of Surgical Oncology},
volume = {50},
number = {1},
abstract = {Introduction: Oesophageal Cancer Multidisciplinary Teams (OC MDTs) operate under significant caseload pressures. This risks variability of decision-making which may influence patient outcomes. Machine Learning (ML) offers the ability to streamline and standardise decision-making by learning from historic treatment decisions to prediction treatment for new patients. We present internally validated ML models designed to predict OC MDT treatment decisions for curative and palliative OC patients.ensuremath<br/ensuremath>ensuremath<br/ensuremath>Methods: four ML algorithms (multinomial logistic regression (MLR), random forests (RF), extreme gradient boost (XGB) and decision tree (DT)) were trained using nested cross-validation on a cohort of 938 OC cases from a single tertiary unit over a 12-year period. The models classified predicted treatments into one of: Surgery (S), Neoadjuvant Chemotherapy (NACT) + S, Neoadjuvant Chemoradiotherapy (NACRT) + S, Endoscopic or Palliative treatment. Performance was assessed on Area Under the Curve (AUC).ensuremath<br/ensuremath>ensuremath<br/ensuremath>Results: across algorithms, all models performed strongly with mean AUC for Surgery = 0.849$±$0.026, NACT +S = 0.884$±$0.008, NACRT +S = 0.834$±$0.035},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Introduction: Oesophageal Cancer Multidisciplinary Teams (OC MDTs) operate under significant caseload pressures. This risks variability of decision-making which may influence patient outcomes. Machine Learning (ML) offers the ability to streamline and standardise decision-making by learning from historic treatment decisions to prediction treatment for new patients. We present internally validated ML models designed to predict OC MDT treatment decisions for curative and palliative OC patients.ensuremath<br/ensuremath>ensuremath<br/ensuremath>Methods: four ML algorithms (multinomial logistic regression (MLR), random forests (RF), extreme gradient boost (XGB) and decision tree (DT)) were trained using nested cross-validation on a cohort of 938 OC cases from a single tertiary unit over a 12-year period. The models classified predicted treatments into one of: Surgery (S), Neoadjuvant Chemotherapy (NACT) + S, Neoadjuvant Chemoradiotherapy (NACRT) + S, Endoscopic or Palliative treatment. Performance was assessed on Area Under the Curve (AUC).ensuremath<br/ensuremath>ensuremath<br/ensuremath>Results: across algorithms, all models performed strongly with mean AUC for Surgery = 0.849$±$0.026, NACT +S = 0.884$±$0.008, NACRT +S = 0.834$±$0.035
Kelly, Thomas Graham; Soorati, Mohammad; Zauner, Klaus-Peter; Ramchurn, Gopal; Tarapore, Danesh
Trade-offs of dynamic control structure in human-swarm systems Proceedings Article
In: The International Symposium on Distributed Autonomous Robotic Systems (DARS) 2024, 2024.
@inproceedings{soton492838,
title = {Trade-offs of dynamic control structure in human-swarm systems},
author = {Thomas Graham Kelly and Mohammad Soorati and Klaus-Peter Zauner and Gopal Ramchurn and Danesh Tarapore},
url = {https://eprints.soton.ac.uk/492838/},
year = {2024},
date = {2024-01-01},
booktitle = {The International Symposium on Distributed Autonomous Robotic Systems (DARS) 2024},
abstract = {Swarm robotics is a study of simple robots that exhibit complex behaviour only by interacting locally with other robots and their environment. The control in swarm robotics is mainly distributed whereas centralised control is widely used in other fields of robotics. Centralised and decentralised control strategies both pose a unique set of benefits and drawbacks for the control of multi-robot systems. While decentralised systems are more scalable and resilient, they are less efficient compared to the centralised systems and they lead to excessive data transmissions to the human operators causing cognitive overload. We examine the trade-offs of each of these approaches in a human-swarm system to perform an environmental monitoring task and propose a flexible hybrid approach, which combines elements of hierarchical and decentralised systems. We find that a flexible hybrid system can outperform a centralised system (in our environmental monitoring task by 19.2%) while reducing the number of messages sent to a human operator (here by 23.1%). We conclude that establishing centralisation for a system is not always optimal for performance and that utilising aspects of centralised and decentralised systems can keep the swarm from hindering its performance.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Swarm robotics is a study of simple robots that exhibit complex behaviour only by interacting locally with other robots and their environment. The control in swarm robotics is mainly distributed whereas centralised control is widely used in other fields of robotics. Centralised and decentralised control strategies both pose a unique set of benefits and drawbacks for the control of multi-robot systems. While decentralised systems are more scalable and resilient, they are less efficient compared to the centralised systems and they lead to excessive data transmissions to the human operators causing cognitive overload. We examine the trade-offs of each of these approaches in a human-swarm system to perform an environmental monitoring task and propose a flexible hybrid approach, which combines elements of hierarchical and decentralised systems. We find that a flexible hybrid system can outperform a centralised system (in our environmental monitoring task by 19.2%) while reducing the number of messages sent to a human operator (here by 23.1%). We conclude that establishing centralisation for a system is not always optimal for performance and that utilising aspects of centralised and decentralised systems can keep the swarm from hindering its performance.
2023
Early, Joseph; Deweese, Ying-Jung Chen; Evers, Christine; Ramchurn, Sarvapali
Extending scene-to-patch models: Multi-resolution multiple instance learning for Earth observation Journal Article
In: Environmental Data Science, vol. 2, pp. 18, 2023.
@article{soton490766,
title = {Extending scene-to-patch models: Multi-resolution multiple instance learning for Earth observation},
author = {Joseph Early and Ying-Jung Chen Deweese and Christine Evers and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/490766/},
year = {2023},
date = {2023-12-01},
journal = {Environmental Data Science},
volume = {2},
pages = {18},
abstract = {Land cover classification (LCC) and natural disaster response (NDR) are important issues in climate change mitigation and adaptation. Existing approaches that use machine learning with Earth observation (EO) imaging data for LCC and NDR often rely on fully annotated and segmented datasets. Creating these datasets requires a large amount of effort, and a lack of suitable datasets has become an obstacle in scaling the use of machine learning for EO. In this study, we extend our prior work on Scene-to-Patch models: an alternative machine learning approach for EO that utilizes Multiple Instance Learning (MIL). As our approach only requires high-level scene labels, it enables much faster development of new datasets while still providing segmentation through patch-level predictions, ultimately increasing the accessibility of using machine learning for EO. We propose new multi-resolution MIL architectures that outperform single-resolution MIL models and non-MIL baselines on the DeepGlobe LCC and FloodNet NDR datasets. In addition, we conduct a thorough analysis of model performance and interpretability.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Land cover classification (LCC) and natural disaster response (NDR) are important issues in climate change mitigation and adaptation. Existing approaches that use machine learning with Earth observation (EO) imaging data for LCC and NDR often rely on fully annotated and segmented datasets. Creating these datasets requires a large amount of effort, and a lack of suitable datasets has become an obstacle in scaling the use of machine learning for EO. In this study, we extend our prior work on Scene-to-Patch models: an alternative machine learning approach for EO that utilizes Multiple Instance Learning (MIL). As our approach only requires high-level scene labels, it enables much faster development of new datasets while still providing segmentation through patch-level predictions, ultimately increasing the accessibility of using machine learning for EO. We propose new multi-resolution MIL architectures that outperform single-resolution MIL models and non-MIL baselines on the DeepGlobe LCC and FloodNet NDR datasets. In addition, we conduct a thorough analysis of model performance and interpretability.
Rigley, Eryn; Bentley, Caitlin; Krook, Joshua; Ramchurn, Gopal
Evaluating international AI skills policy: a systematic review of AI skills policy in seven countries Journal Article
In: Global Policy, 2023, (Funding Information: This research was supported via UKRI by the DCMS Science and Analysis R&D Programme. It was developed and produced according to UKRI's initial hypotheses and output requests. Any primary research, subsequent findings or recommendations do not represent Government views or policy and are produced according to academic ethics, quality assurance and independence.).
@article{soton485727,
title = {Evaluating international AI skills policy: a systematic review of AI skills policy in seven countries},
author = {Eryn Rigley and Caitlin Bentley and Joshua Krook and Gopal Ramchurn},
url = {https://eprints.soton.ac.uk/485727/},
year = {2023},
date = {2023-12-01},
journal = {Global Policy},
abstract = {ensuremath<pensuremath>As artificial intelligence (AI) is having an increasingly disruptive impact across industries, companies continue to report having difficulty when recruiting for AI roles, while new graduates find it difficult to find employment, indicating a skills gap or skills misalignment. International approaches to AI skills programmes can offer a guide to future policy development of a skilled workforce, best placed to harness the economic opportunities that AI may support. The authors performed a systematic literature review on AI skills in government policies and documents from seven countries: Australia, Canada, China, Singapore, Sweden, the United Kingom and the United States. We found a divide between countries which emphasised a broader, nationwide approach to upskill and educate all citizens at different levels, namely the United States and Singapore and those countries which emphasised a narrower focus on educating a smaller group of experts with advanced AI knowledge and skills, namely China, Sweden and Canada. We found that the former, broader approaches tended to correlate with higher AI readiness and index scores than the narrower, expert-driven approach. Our findings indicate that, to match world-leading AI readiness, future AI skills policy should follow these broad, nationwide approaches to upskill and educate all citizens at different levels of AI expertise.ensuremath</pensuremath>},
note = {Funding Information:
This research was supported via UKRI by the DCMS Science and Analysis R&D Programme. It was developed and produced according to UKRI's initial hypotheses and output requests. Any primary research, subsequent findings or recommendations do not represent Government views or policy and are produced according to academic ethics, quality assurance and independence.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>As artificial intelligence (AI) is having an increasingly disruptive impact across industries, companies continue to report having difficulty when recruiting for AI roles, while new graduates find it difficult to find employment, indicating a skills gap or skills misalignment. International approaches to AI skills programmes can offer a guide to future policy development of a skilled workforce, best placed to harness the economic opportunities that AI may support. The authors performed a systematic literature review on AI skills in government policies and documents from seven countries: Australia, Canada, China, Singapore, Sweden, the United Kingom and the United States. We found a divide between countries which emphasised a broader, nationwide approach to upskill and educate all citizens at different levels, namely the United States and Singapore and those countries which emphasised a narrower focus on educating a smaller group of experts with advanced AI knowledge and skills, namely China, Sweden and Canada. We found that the former, broader approaches tended to correlate with higher AI readiness and index scores than the narrower, expert-driven approach. Our findings indicate that, to match world-leading AI readiness, future AI skills policy should follow these broad, nationwide approaches to upskill and educate all citizens at different levels of AI expertise.ensuremath</pensuremath>
Singh, Lokesh; Ramchurn, Gopal
The effect of automated agents on individual performance under induced stress Proceedings Article
In: Kalra, Jay (Ed.): Emerging Technologies in Healthcare and Medicine: Proceedings of the AHFE International Conference on Human Factors in Design, Engineering and Computing (AHFE 2023 Hawaii Edition), pp. 118–127, AHFE International, 2023.
@inproceedings{soton485655,
title = {The effect of automated agents on individual performance under induced stress},
author = {Lokesh Singh and Gopal Ramchurn},
editor = {Jay Kalra},
url = {https://eprints.soton.ac.uk/485655/},
year = {2023},
date = {2023-11-01},
booktitle = {Emerging Technologies in Healthcare and Medicine: Proceedings of the AHFE International Conference on Human Factors in Design, Engineering and Computing (AHFE 2023 Hawaii Edition)},
pages = {118–127},
publisher = {AHFE International},
abstract = {Induced stress is a phenomenon commonly experienced across different fields such as emergency services, healthcare, air traffic control, sports, and business - which necessitates the development of effective coping strategies and resilience for individuals or teams performing under pressure. This study aims to examine the effects of automated agents on individual performance during high-stress conditions. The design of these agents ensures they carry out identical tasks as participants based on predetermined frameworks. Participants underwent an experimentally designed task that aimed at inducing stress while measuring their performance amidst time pressure and auditory distraction. Results indicate that working with automated agents causes individuals to alter their approach by focusing narrowly on immediate concerns - making it challenging for them to consider several options or see broader contexts accurately. Regardless of ability level participants' performances were influenced by these automated agents. Future research will explore how these findings interact with physiological signals. This study highlights the importance of developing effective coping strategies and the potential impact of social factors on individual performance under induced stress.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Induced stress is a phenomenon commonly experienced across different fields such as emergency services, healthcare, air traffic control, sports, and business - which necessitates the development of effective coping strategies and resilience for individuals or teams performing under pressure. This study aims to examine the effects of automated agents on individual performance during high-stress conditions. The design of these agents ensures they carry out identical tasks as participants based on predetermined frameworks. Participants underwent an experimentally designed task that aimed at inducing stress while measuring their performance amidst time pressure and auditory distraction. Results indicate that working with automated agents causes individuals to alter their approach by focusing narrowly on immediate concerns - making it challenging for them to consider several options or see broader contexts accurately. Regardless of ability level participants' performances were influenced by these automated agents. Future research will explore how these findings interact with physiological signals. This study highlights the importance of developing effective coping strategies and the potential impact of social factors on individual performance under induced stress.
Thavanesan, Navamayooran; Bodala, Indu; Walters, Zoe; Ramchurn, Sarvapali; Underwood, Timothy J.; Vigneswaran, Ganesh
Machine learning to predict curative multidisciplinary team treatment decisions in oesophageal cancer Journal Article
In: European Journal of Surgical Oncology, vol. 49, no. 11, 2023, (Publisher Copyright: © 2023 The Author(s)).
@article{soton479497b,
title = {Machine learning to predict curative multidisciplinary team treatment decisions in oesophageal cancer},
author = {Navamayooran Thavanesan and Indu Bodala and Zoe Walters and Sarvapali Ramchurn and Timothy J. Underwood and Ganesh Vigneswaran},
url = {https://eprints.soton.ac.uk/479497/},
year = {2023},
date = {2023-11-01},
journal = {European Journal of Surgical Oncology},
volume = {49},
number = {11},
abstract = {ensuremath<pensuremath>Background: Rising workflow pressures within the oesophageal cancer (OC) multidisciplinary team (MDT) can lead to variability in decision-making, and health inequality. Machine learning (ML) offers a potential automated data-driven approach to address inconsistency and standardize care. The aim of this experimental pilot study was to develop ML models able to predict curative OC MDT treatment decisions and determine the relative importance of underlying decision-critical variables. Methods: Retrospective complete-case analysis of oesophagectomy patients $±$ neoadjuvant chemotherapy (NACT) or chemoradiotherapy (NACRT) between 2010 and 2020. Established ML algorithms (Multinomial Logistic regression (MLR), Random Forests (RF), Extreme Gradient Boosting (XGB)) and Decision Tree (DT) were used to train models predicting OC MDT treatment decisions: surgery (S), NACT + S or NACRT + S. Performance metrics included Area Under the Curve (AUC), Accuracy, Kappa, LogLoss, F1 and Precision -Recall AUC. Variable importance was calculated for each model. Results: We identified 399 cases with a male-to-female ratio of 3.6:1 and median age of 66.1yrs (range 32?83). MLR outperformed RF, XGB and DT across performance metrics (mean AUC of 0.793 [$±$0.045] vs 0.757 [$±$0.068], 0.740 [$±$0.042], and 0.709 [$±$0.021] respectively). Variable importance analysis identified age as a major factor in the decision to offer surgery alone or NACT + S across models (p < 0.05). Conclusions: ML techniques can use limited feature-sets to predict curative UGI MDT treatment decisions. Explainable Artificial Intelligence methods provide insight into decision-critical variables, highlighting underlying subconscious biases in cancer care decision-making. Such models may allow prioritization of caseload, improve efficiency, and offer data-driven decision-assistance to MDTs in the future.ensuremath</pensuremath>},
note = {Publisher Copyright:
© 2023 The Author(s)},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Background: Rising workflow pressures within the oesophageal cancer (OC) multidisciplinary team (MDT) can lead to variability in decision-making, and health inequality. Machine learning (ML) offers a potential automated data-driven approach to address inconsistency and standardize care. The aim of this experimental pilot study was to develop ML models able to predict curative OC MDT treatment decisions and determine the relative importance of underlying decision-critical variables. Methods: Retrospective complete-case analysis of oesophagectomy patients $±$ neoadjuvant chemotherapy (NACT) or chemoradiotherapy (NACRT) between 2010 and 2020. Established ML algorithms (Multinomial Logistic regression (MLR), Random Forests (RF), Extreme Gradient Boosting (XGB)) and Decision Tree (DT) were used to train models predicting OC MDT treatment decisions: surgery (S), NACT + S or NACRT + S. Performance metrics included Area Under the Curve (AUC), Accuracy, Kappa, LogLoss, F1 and Precision -Recall AUC. Variable importance was calculated for each model. Results: We identified 399 cases with a male-to-female ratio of 3.6:1 and median age of 66.1yrs (range 32?83). MLR outperformed RF, XGB and DT across performance metrics (mean AUC of 0.793 [$±$0.045] vs 0.757 [$±$0.068], 0.740 [$±$0.042], and 0.709 [$±$0.021] respectively). Variable importance analysis identified age as a major factor in the decision to offer surgery alone or NACT + S across models (p < 0.05). Conclusions: ML techniques can use limited feature-sets to predict curative UGI MDT treatment decisions. Explainable Artificial Intelligence methods provide insight into decision-critical variables, highlighting underlying subconscious biases in cancer care decision-making. Such models may allow prioritization of caseload, improve efficiency, and offer data-driven decision-assistance to MDTs in the future.ensuremath</pensuremath>
Krook, Joshua; Williams, Jennifer; Seabrooke, Tina; Schneiders, Eike; Blockx, Jan; Middleton, Stuart E; Ramchurn, Sarvapali
AI large language models inquiry: TASHub Response Miscellaneous
2023.
@misc{soton481740,
title = {AI large language models inquiry: TASHub Response},
author = {Joshua Krook and Jennifer Williams and Tina Seabrooke and Eike Schneiders and Jan Blockx and Stuart E Middleton and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/481740/},
year = {2023},
date = {2023-08-01},
publisher = {University of Southampton},
abstract = {Policy submission to the Consultation by Communications and Digital Committee, House of Lords, AI Large Language Models Inquiry.ensuremath<br/ensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Policy submission to the Consultation by Communications and Digital Committee, House of Lords, AI Large Language Models Inquiry.ensuremath<br/ensuremath>
Krook, Joshua; Williams, Jennifer; Seabrooke, Tina; Schneiders, Eike; Blockx, Jan; Middleton, Stuart E; Ramchurn, Sarvapali
AI large language models inquiry: TASHub response Miscellaneous
2023.
@misc{soton481740b,
title = {AI large language models inquiry: TASHub response},
author = {Joshua Krook and Jennifer Williams and Tina Seabrooke and Eike Schneiders and Jan Blockx and Stuart E Middleton and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/481740/},
year = {2023},
date = {2023-08-01},
publisher = {University of Southampton},
abstract = {Policy submission to the Consultation by Communications and Digital Committee, House of Lords, AI Large Language Models Inquiry.ensuremath<br/ensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Policy submission to the Consultation by Communications and Digital Committee, House of Lords, AI Large Language Models Inquiry.ensuremath<br/ensuremath>
Thavanesan, Navamayooran; Bodala, Indu; Walters, Zoe; Ramchurn, Sarvapali; Underwood, Timothy J; Vigneswaran, Ganesh
Machine learning to predict curative multidisciplinary team treatment decisions in oesophageal cancer Journal Article
In: European Journal of Surgical Oncology, 2023, (Publisher Copyright: copyright 2023 The Author(s)).
@article{soton479497,
title = {Machine learning to predict curative multidisciplinary team treatment decisions in oesophageal cancer},
author = {Navamayooran Thavanesan and Indu Bodala and Zoe Walters and Sarvapali Ramchurn and Timothy J Underwood and Ganesh Vigneswaran},
url = {https://eprints.soton.ac.uk/479497/},
year = {2023},
date = {2023-07-01},
journal = {European Journal of Surgical Oncology},
abstract = {ensuremath<pensuremath>Background: Rising workflow pressures within the oesophageal cancer (OC) multidisciplinary team (MDT) can lead to variability in decision-making, and health inequality. Machine learning (ML) offers a potential automated data-driven approach to address inconsistency and standardize care. The aim of this experimental pilot study was to develop ML models able to predict curative OC MDT treatment decisions and determine the relative importance of underlying decision-critical variables. Methods: Retrospective complete-case analysis of oesophagectomy patients $pm$ neoadjuvant chemotherapy (NACT) or chemoradiotherapy (NACRT) between 2010 and 2020. Established ML algorithms (Multinomial Logistic regression (MLR), Random Forests (RF), Extreme Gradient Boosting (XGB)) and Decision Tree (DT) were used to train models predicting OC MDT treatment decisions: surgery (S), NACT + S or NACRT + S. Performance metrics included Area Under the Curve (AUC), Accuracy, Kappa, LogLoss, F1 and Precision -Recall AUC. Variable importance was calculated for each model. Results: We identified 399 cases with a male-to-female ratio of 3.6:1 and median age of 66.1yrs (range 32?83). MLR outperformed RF, XGB and DT across performance metrics (mean AUC of 0.793 [$pm$0.045] vs 0.757 [$pm$0.068], 0.740 [$pm$0.042], and 0.709 [$pm$0.021] respectively). Variable importance analysis identified age as a major factor in the decision to offer surgery alone or NACT + S across models (p < 0.05). Conclusions: ML techniques can use limited feature-sets to predict curative UGI MDT treatment decisions. Explainable Artificial Intelligence methods provide insight into decision-critical variables, highlighting underlying subconscious biases in cancer care decision-making. Such models may allow prioritization of caseload, improve efficiency, and offer data-driven decision-assistance to MDTs in the future.ensuremath</pensuremath>},
note = {Publisher Copyright:
copyright 2023 The Author(s)},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Background: Rising workflow pressures within the oesophageal cancer (OC) multidisciplinary team (MDT) can lead to variability in decision-making, and health inequality. Machine learning (ML) offers a potential automated data-driven approach to address inconsistency and standardize care. The aim of this experimental pilot study was to develop ML models able to predict curative OC MDT treatment decisions and determine the relative importance of underlying decision-critical variables. Methods: Retrospective complete-case analysis of oesophagectomy patients $pm$ neoadjuvant chemotherapy (NACT) or chemoradiotherapy (NACRT) between 2010 and 2020. Established ML algorithms (Multinomial Logistic regression (MLR), Random Forests (RF), Extreme Gradient Boosting (XGB)) and Decision Tree (DT) were used to train models predicting OC MDT treatment decisions: surgery (S), NACT + S or NACRT + S. Performance metrics included Area Under the Curve (AUC), Accuracy, Kappa, LogLoss, F1 and Precision -Recall AUC. Variable importance was calculated for each model. Results: We identified 399 cases with a male-to-female ratio of 3.6:1 and median age of 66.1yrs (range 32?83). MLR outperformed RF, XGB and DT across performance metrics (mean AUC of 0.793 [$pm$0.045] vs 0.757 [$pm$0.068], 0.740 [$pm$0.042], and 0.709 [$pm$0.021] respectively). Variable importance analysis identified age as a major factor in the decision to offer surgery alone or NACT + S across models (p < 0.05). Conclusions: ML techniques can use limited feature-sets to predict curative UGI MDT treatment decisions. Explainable Artificial Intelligence methods provide insight into decision-critical variables, highlighting underlying subconscious biases in cancer care decision-making. Such models may allow prioritization of caseload, improve efficiency, and offer data-driven decision-assistance to MDTs in the future.ensuremath</pensuremath>
Abioye, Ayodeji
University of Southampton, 2023.
@phdthesis{soton479472,
title = {Multimodal speech and visual gesture control interface technique for small unmanned multirotor aircraft},
author = {Ayodeji Abioye},
url = {https://eprints.soton.ac.uk/479472/},
year = {2023},
date = {2023-07-01},
publisher = {University of Southampton},
school = {University of Southampton},
abstract = {ensuremath<p class="MsoNormal"ensuremath>This research conducted an investigation into the use of novel human computer interaction(HCI) interfaces in the control of small multirotor unmanned aerial vehicles(UAVs). The main objective was to propose, design, and develop an alternative control interface for the small multirotor UAV, which could perform better than the standard RC joystick (RCJ) controller, and to evaluate the performance of the proposed interface. The multimodal speech and visual gesture (mSVG)interface were proposed, designed, and developed. This was then coupled to a Rotor S ROS Gazebo UAV simulator. An experiment study was designed to determine how practical the use of the proposed multimodal speech and visual gesture interface was in the control of small multirotor UAVs by determining the limits of speech and gesture at different ambient noise levels and under different background-lighting conditions, respectively. And to determine how the mSVG interface compares to the RC joystick controller for a simple navigational control task - in terms of performance (time of completion and accuracy of navigational control) and from a human factor?s perspective (user satisfaction and cognitive workload). 37 participants were recruited. From the results of the experiments conducted, the mSVG interface was found to be an effective alternative to the RCJ interface when operated within a constrained application environment. From the result of the noise level experiment, it was observed that speech recognition accuracy/success rate falls as noise levels rise, with75 dB noise level being the practical aerial robot (aerobot) application limit. From the results of the gesture lighting experiment, gestures were successfully recognised from 10 Lux and above on distinct solid backgrounds, but the effect of varying both the lighting conditions and the environment background on the quality of gesture recognition, was insignificant (< 0.5%), implying that the technology used, type of gesture captured, and the image processing technique used were more important. From the result of the performance and cognitive workload comparison between the RCJ and mSVG interfaces, the mSVG interface was found to perform better at higher nCA application levels than the RCJ interface. The mSVG interface was 1 minute faster and 25% more accurate than the RCJ interface; and the RCJ interface was found to be 1.4 times more cognitively demanding than the mSVG interface. The main limitation of this research was the limited lighting level range of 10 Lux - 1400 Lux used during the gesture lighting experiment, which constrains the application limit to lowlighting indoor environments. Suggested further works from this research included the development of a more robust gesture and speech algorithm and the coupling of the improved mSVG interface on to a practical UAV.ensuremath</pensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {phdthesis}
}
ensuremath<p class="MsoNormal"ensuremath>This research conducted an investigation into the use of novel human computer interaction(HCI) interfaces in the control of small multirotor unmanned aerial vehicles(UAVs). The main objective was to propose, design, and develop an alternative control interface for the small multirotor UAV, which could perform better than the standard RC joystick (RCJ) controller, and to evaluate the performance of the proposed interface. The multimodal speech and visual gesture (mSVG)interface were proposed, designed, and developed. This was then coupled to a Rotor S ROS Gazebo UAV simulator. An experiment study was designed to determine how practical the use of the proposed multimodal speech and visual gesture interface was in the control of small multirotor UAVs by determining the limits of speech and gesture at different ambient noise levels and under different background-lighting conditions, respectively. And to determine how the mSVG interface compares to the RC joystick controller for a simple navigational control task - in terms of performance (time of completion and accuracy of navigational control) and from a human factor?s perspective (user satisfaction and cognitive workload). 37 participants were recruited. From the results of the experiments conducted, the mSVG interface was found to be an effective alternative to the RCJ interface when operated within a constrained application environment. From the result of the noise level experiment, it was observed that speech recognition accuracy/success rate falls as noise levels rise, with75 dB noise level being the practical aerial robot (aerobot) application limit. From the results of the gesture lighting experiment, gestures were successfully recognised from 10 Lux and above on distinct solid backgrounds, but the effect of varying both the lighting conditions and the environment background on the quality of gesture recognition, was insignificant (< 0.5%), implying that the technology used, type of gesture captured, and the image processing technique used were more important. From the result of the performance and cognitive workload comparison between the RCJ and mSVG interfaces, the mSVG interface was found to perform better at higher nCA application levels than the RCJ interface. The mSVG interface was 1 minute faster and 25% more accurate than the RCJ interface; and the RCJ interface was found to be 1.4 times more cognitively demanding than the mSVG interface. The main limitation of this research was the limited lighting level range of 10 Lux - 1400 Lux used during the gesture lighting experiment, which constrains the application limit to lowlighting indoor environments. Suggested further works from this research included the development of a more robust gesture and speech algorithm and the coupling of the improved mSVG interface on to a practical UAV.ensuremath</pensuremath>
Abioye, Ayodeji; Naiseh, Mohammad; Hunt, William; Clark, Jediah R; Ramchurn, Sarvapali D; Soorati, Mohammad
The effect of data visualisation quality and task density on human-swarm interaction Proceedings Article
In: Proceedings of the 2023 32nd IEEE International Conference on Robot and Human Interactive Communication (RO-MAN), IEEE, 2023.
@inproceedings{soton479970,
title = {The effect of data visualisation quality and task density on human-swarm interaction},
author = {Ayodeji Abioye and Mohammad Naiseh and William Hunt and Jediah R Clark and Sarvapali D Ramchurn and Mohammad Soorati},
url = {https://eprints.soton.ac.uk/479970/},
year = {2023},
date = {2023-06-01},
urldate = {2023-06-01},
booktitle = {Proceedings of the 2023 32nd IEEE International Conference on Robot and Human Interactive Communication (RO-MAN)},
publisher = {IEEE},
abstract = {Despite the advantages of having robot swarms, human supervision is required for real-world applications. The performance of the human-swarm system depends on several factors including the data availability for the human operators. In this paper, we study the human factors aspect of the human-swarm interaction and investigate how having access to high-quality data can affect the performance of the human-swarm system - the number of tasks completed and the human trust level in operation. We designed an experiment where a human operator is tasked to operate a swarm to identify casualties in an area within a given time period. One group of operators had the option to request high-quality pictures while the other group had to base their decision on the available low-quality images. We performed a user study with 120 participants and recorded their success rate (directly logged via the simulation platform) as well as their workload and trust level (measured through a questionnaire after completing a human-swarm scenario). The findings from our study indicated that the group granted access to high-quality data exhibited an increased workload and placed greater trust in the swarm, thus confirming our initial hypothesis. However, we also found that the number of accurately identified casualties did not significantly vary between the two groups, suggesting that data quality had no impact on the successful completion of tasks.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Despite the advantages of having robot swarms, human supervision is required for real-world applications. The performance of the human-swarm system depends on several factors including the data availability for the human operators. In this paper, we study the human factors aspect of the human-swarm interaction and investigate how having access to high-quality data can affect the performance of the human-swarm system - the number of tasks completed and the human trust level in operation. We designed an experiment where a human operator is tasked to operate a swarm to identify casualties in an area within a given time period. One group of operators had the option to request high-quality pictures while the other group had to base their decision on the available low-quality images. We performed a user study with 120 participants and recorded their success rate (directly logged via the simulation platform) as well as their workload and trust level (measured through a questionnaire after completing a human-swarm scenario). The findings from our study indicated that the group granted access to high-quality data exhibited an increased workload and placed greater trust in the swarm, thus confirming our initial hypothesis. However, we also found that the number of accurately identified casualties did not significantly vary between the two groups, suggesting that data quality had no impact on the successful completion of tasks.
Krook, Joshua; McAuley, Derek; Anderson, Stuart; Downer, John; Winter, Peter; Ramchurn, Sarvapali D
AI Foundation Models: initial review, CMA Consultation, TAS Hub Response Miscellaneous
2023.
@misc{soton477553,
title = {AI Foundation Models: initial review, CMA Consultation, TAS Hub Response},
author = {Joshua Krook and Derek McAuley and Stuart Anderson and John Downer and Peter Winter and Sarvapali D Ramchurn},
url = {https://eprints.soton.ac.uk/477553/},
year = {2023},
date = {2023-06-01},
urldate = {2023-06-01},
publisher = {University of Southampton},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Krook, Joshua; Downer, John; Winter, Peter; Williams, Jennifer; Ives, Jonathan; Bratu, Roxana; Sheir, Stephanie; Williams, Robin; Anderson, Stuart; Li, Phoebe; Ramamoorthy, Subramanian; Ramchurn, Sarvapali
AI regulation: a pro-innovation approach ? policy proposals: TASHub Response Miscellaneous
2023.
@misc{soton478329,
title = {AI regulation: a pro-innovation approach ? policy proposals: TASHub Response},
author = {Joshua Krook and John Downer and Peter Winter and Jennifer Williams and Jonathan Ives and Roxana Bratu and Stephanie Sheir and Robin Williams and Stuart Anderson and Phoebe Li and Subramanian Ramamoorthy and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/478329/},
year = {2023},
date = {2023-06-01},
publisher = {University of Southampton},
abstract = {Response to open consultation from: Department for Science, Innovation and Technologyensuremath<br/ensuremath>and Office for Artificial Intelligence},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Response to open consultation from: Department for Science, Innovation and Technologyensuremath<br/ensuremath>and Office for Artificial Intelligence
Hunt, William; Ryan, Jack; Abioye, Ayodeji O; Ramchurn, Sarvapali D; Soorati, Mohammad D
Demonstrating performance benefits of human-swarm teaming Proceedings Article
In: Proceedings of the 2023 International Conference on Autonomous Agents and Multiagent Systems, pp. 3062–3064, International Foundation for Autonomous Agents and Multiagent Systems (IFAAMAS), 2023.
@inproceedings{soton479903,
title = {Demonstrating performance benefits of human-swarm teaming},
author = {William Hunt and Jack Ryan and Ayodeji O Abioye and Sarvapali D Ramchurn and Mohammad D Soorati},
url = {https://eprints.soton.ac.uk/479903/},
year = {2023},
date = {2023-05-01},
urldate = {2023-05-01},
booktitle = {Proceedings of the 2023 International Conference on Autonomous Agents and Multiagent Systems},
pages = {3062–3064},
publisher = {International Foundation for Autonomous Agents and Multiagent Systems (IFAAMAS)},
abstract = {Autonomous swarms of robots can bring robustness, scalability and adaptability to safety-critical tasks such as search and rescue but their application is still very limited. Using semi-autonomous swarms with human control can bring robot swarms to real-world applications. Human operators can define goals for the swarm, monitor their performance and interfere with, or overrule, the decisions and behaviour. We present the "Human And Robot Interactive Swarm'' simulator (HARIS) that allows multi-user interaction with a robot swarm and facilitates qualitative and quantitative user studies through simulation of robot swarms completing tasks, from package delivery to search and rescue, with varying levels of human control. In this demonstration, we showcase the simulator by using it to study the performance gain offered by maintaining a "human-in-the-loop'' over a fully autonomous system as an example. This is illustrated in the context of search and rescue, with an autonomous allocation of resources to those in need.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Autonomous swarms of robots can bring robustness, scalability and adaptability to safety-critical tasks such as search and rescue but their application is still very limited. Using semi-autonomous swarms with human control can bring robot swarms to real-world applications. Human operators can define goals for the swarm, monitor their performance and interfere with, or overrule, the decisions and behaviour. We present the "Human And Robot Interactive Swarm'' simulator (HARIS) that allows multi-user interaction with a robot swarm and facilitates qualitative and quantitative user studies through simulation of robot swarms completing tasks, from package delivery to search and rescue, with varying levels of human control. In this demonstration, we showcase the simulator by using it to study the performance gain offered by maintaining a "human-in-the-loop'' over a fully autonomous system as an example. This is illustrated in the context of search and rescue, with an autonomous allocation of resources to those in need.
Worrawichaipat, Phuriwat; Gerding, Enrico; Kaparias, Ioannis; Ramchurn, Sarvapali
Multi-agent signal-less intersection management with dynamic platoon formation Proceedings Article
In: 22nd International Conference on Autonomous Agents and Multiagent Systems (29/05/23 - 02/06/23), pp. 1542–1550, 2023.
@inproceedings{soton478647,
title = {Multi-agent signal-less intersection management with dynamic platoon formation},
author = {Phuriwat Worrawichaipat and Enrico Gerding and Ioannis Kaparias and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/478647/},
year = {2023},
date = {2023-05-01},
urldate = {2023-05-01},
booktitle = {22nd International Conference on Autonomous Agents and Multiagent Systems (29/05/23 - 02/06/23)},
pages = {1542–1550},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Everett, Gregory; Beal, Ryan J; Matthews, Tim; Early, Joseph; Norman, Timothy J; Ramchurn, Sarvapali D
Inferring player location in sports matches: multi-agent spatial imputation from limited observations Miscellaneous
2023.
@misc{soton477020,
title = {Inferring player location in sports matches: multi-agent spatial imputation from limited observations},
author = {Gregory Everett and Ryan J Beal and Tim Matthews and Joseph Early and Timothy J Norman and Sarvapali D Ramchurn},
url = {https://eprints.soton.ac.uk/477020/},
year = {2023},
date = {2023-02-01},
abstract = {Understanding agent behaviour in Multi-Agent Systems (MAS) is an important problem in domains such as autonomous driving, disaster response, and sports analytics. Existing MAS problems typically use uniform timesteps with observations for all agents. In this work, we analyse the problem of agent location imputation, specifically posed in environments with non-uniform timesteps and limited agent observability (textttchar12695% missing values). Our approach uses Long Short-Term Memory and Graph Neural Network components to learn temporal and inter-agent patterns to predict the location of all agents at every timestep. We apply this to the domain of football (soccer) by imputing the location of all players in a game from sparse event data (e.g., shots and passes). Our model estimates player locations to within textttchar1266.9m; a textttchar12662% reduction in error from the best performing baseline. This approach facilitates downstream analysis tasks such as player physical metrics, player coverage, and team pitch control. Existing solutions to these tasks often require optical tracking data, which is expensive to obtain and only available to elite clubs. By imputing player locations from easy to obtain event data, we increase the accessibility of downstream tasks.},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Understanding agent behaviour in Multi-Agent Systems (MAS) is an important problem in domains such as autonomous driving, disaster response, and sports analytics. Existing MAS problems typically use uniform timesteps with observations for all agents. In this work, we analyse the problem of agent location imputation, specifically posed in environments with non-uniform timesteps and limited agent observability (textttchar12695% missing values). Our approach uses Long Short-Term Memory and Graph Neural Network components to learn temporal and inter-agent patterns to predict the location of all agents at every timestep. We apply this to the domain of football (soccer) by imputing the location of all players in a game from sparse event data (e.g., shots and passes). Our model estimates player locations to within textttchar1266.9m; a textttchar12662% reduction in error from the best performing baseline. This approach facilitates downstream analysis tasks such as player physical metrics, player coverage, and team pitch control. Existing solutions to these tasks often require optical tracking data, which is expensive to obtain and only available to elite clubs. By imputing player locations from easy to obtain event data, we increase the accessibility of downstream tasks.
2022
Ahmed, Sarah; Azim, Tayyaba; Early, Joseph Arthur; Ramchurn, Sarvapali
Revisiting deep fisher vectors: using fisher information to improve object classification Proceedings Article
In: British Machine Vision Conference (21/11/22 - 24/11/22), 2022.
@inproceedings{soton471260,
title = {Revisiting deep fisher vectors: using fisher information to improve object classification},
author = {Sarah Ahmed and Tayyaba Azim and Joseph Arthur Early and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/471260/},
year = {2022},
date = {2022-11-01},
booktitle = {British Machine Vision Conference (21/11/22 - 24/11/22)},
abstract = {Although deep learning models have become the gold standard in achieving outstanding results on a large variety of computer vision and machine learning tasks, the use of kernel methods has still not gone out of trend because of its potential to beat deep learning performances at a number of occasions. Given the potential of kernel techniques, prior works have also proposed the use of hybrid approaches combining deep learning with kernel learning to complement their respective strengths and weaknesses. This work develops this idea further by introducing an improved version of Fisher kernels derived from the deep Boltzmann machines (DBM). Our improved deep Fisher kernel (IDFK) utilises an approximation of the Fisher information matrix to derive improved Fisher vectors. We show IDFK can be utilised to retain a high degree of class separability, making it appropriate for classification and retrieval tasks. The efficacy of the proposed approach is evaluated on three benchmark data sets: MNIST, USPS and Alphanumeric, showing an improvement in classification performance over existing kernel approaches, and comparable performance to deep learning methods, but with much reduced computational costs. Using explainable AI methods, we also demonstrate why our IDFK leads to better classification performance.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Although deep learning models have become the gold standard in achieving outstanding results on a large variety of computer vision and machine learning tasks, the use of kernel methods has still not gone out of trend because of its potential to beat deep learning performances at a number of occasions. Given the potential of kernel techniques, prior works have also proposed the use of hybrid approaches combining deep learning with kernel learning to complement their respective strengths and weaknesses. This work develops this idea further by introducing an improved version of Fisher kernels derived from the deep Boltzmann machines (DBM). Our improved deep Fisher kernel (IDFK) utilises an approximation of the Fisher information matrix to derive improved Fisher vectors. We show IDFK can be utilised to retain a high degree of class separability, making it appropriate for classification and retrieval tasks. The efficacy of the proposed approach is evaluated on three benchmark data sets: MNIST, USPS and Alphanumeric, showing an improvement in classification performance over existing kernel approaches, and comparable performance to deep learning methods, but with much reduced computational costs. Using explainable AI methods, we also demonstrate why our IDFK leads to better classification performance.
Yazdanpanah, Vahid; Gerding, Enrico; Stein, Sebastian; Dastani, Mehdi; Jonker, Catholijn M; Norman, Timothy; Ramchurn, Sarvapali
Reasoning About Responsibility in Autonomous Systems: Challenges and Opportunities Journal Article
In: AI & Society, 2022.
@article{soton471971,
title = {Reasoning About Responsibility in Autonomous Systems: Challenges and Opportunities},
author = {Vahid Yazdanpanah and Enrico Gerding and Sebastian Stein and Mehdi Dastani and Catholijn M Jonker and Timothy Norman and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/471971/},
year = {2022},
date = {2022-11-01},
journal = {AI & Society},
abstract = {Ensuring the trustworthiness of autonomous systems and artificial intelligenceensuremath<br/ensuremath>is an important interdisciplinary endeavour. In this position paper, we argue thatensuremath<br/ensuremath>this endeavour will benefit from technical advancements in capturing various forms of responsibility, and we present a comprehensive research agenda to achieve this. In particular, we argue that ensuring the reliability of autonomous system can take advantage of technical approaches for quantifying degrees of responsibility and for coordinating tasks based on that. Moreover, we deem that, in certifying the legality of an AI system, formal and computationally implementable notions of responsibility, blame, accountability, and liability are applicable for addressing potential responsibility gaps (i.e., situations in which a group is responsible, but individuals? responsibility may be unclear). This is a call to enable AI systems themselves, as well as those involved in the design, monitoring, and governance of AI systems, to represent and reason about who can be seen as responsible in prospect (e.g., for completing a task in future) and who can be seen as responsible retrospectively (e.g., for a failure that has already occurred). To that end, in this work, we show that across all stages of the design, development, and deployment of Trustworthy Autonomous Systems (TAS), responsibility reasoning should play a key role. This position paper is the first step towards establishing a road-map and research agenda on how the notion of responsibility can provide novel solution concepts for ensuring the reliability and legality of TAS and, as a result, enables an effective embedding of AI technologies into society.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Ensuring the trustworthiness of autonomous systems and artificial intelligenceensuremath<br/ensuremath>is an important interdisciplinary endeavour. In this position paper, we argue thatensuremath<br/ensuremath>this endeavour will benefit from technical advancements in capturing various forms of responsibility, and we present a comprehensive research agenda to achieve this. In particular, we argue that ensuring the reliability of autonomous system can take advantage of technical approaches for quantifying degrees of responsibility and for coordinating tasks based on that. Moreover, we deem that, in certifying the legality of an AI system, formal and computationally implementable notions of responsibility, blame, accountability, and liability are applicable for addressing potential responsibility gaps (i.e., situations in which a group is responsible, but individuals? responsibility may be unclear). This is a call to enable AI systems themselves, as well as those involved in the design, monitoring, and governance of AI systems, to represent and reason about who can be seen as responsible in prospect (e.g., for completing a task in future) and who can be seen as responsible retrospectively (e.g., for a failure that has already occurred). To that end, in this work, we show that across all stages of the design, development, and deployment of Trustworthy Autonomous Systems (TAS), responsibility reasoning should play a key role. This position paper is the first step towards establishing a road-map and research agenda on how the notion of responsibility can provide novel solution concepts for ensuring the reliability and legality of TAS and, as a result, enables an effective embedding of AI technologies into society.
Early, Joseph; Deweese, Ying-Jung; Evers, Christine; Ramchurn, Sarvapali
Scene-to-Patch earth observation: multiple instance learning for land cover classification Miscellaneous
2022, (14 pages total (4 main content; 2 acknowledgments + citations; 8 appendices); 8 figures (2 main; 6 appendix); published at "Tackling Climate Change with Machine Learning: Workshop at NeurIPS 2022").
@misc{soton472853,
title = {Scene-to-Patch earth observation: multiple instance learning for land cover classification},
author = {Joseph Early and Ying-Jung Deweese and Christine Evers and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/472853/},
year = {2022},
date = {2022-11-01},
abstract = {Land cover classification (LCC), and monitoring how land use changes over time, is an important process in climate change mitigation and adaptation. Existing approaches that use machine learning with Earth observation data for LCC rely on fully-annotated and segmented datasets. Creating these datasets requires a large amount of effort, and a lack of suitable datasets has become an obstacle in scaling the use of LCC. In this study, we propose Scene-to-Patch models: an alternative LCC approach utilising Multiple Instance Learning (MIL) that requires only high-level scene labels. This enables much faster development of new datasets whilst still providing segmentation through patch-level predictions, ultimately increasing the accessibility of using LCC for different scenarios. On the DeepGlobe-LCC dataset, our approach outperforms non-MIL baselines on both scene- and patch-level prediction. This work provides the foundation for expanding the use of LCC in climate change mitigation methods for technology, government, and academia.},
note = {14 pages total (4 main content; 2 acknowledgments + citations; 8 appendices); 8 figures (2 main; 6 appendix); published at "Tackling Climate Change with Machine Learning: Workshop at NeurIPS 2022"},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Land cover classification (LCC), and monitoring how land use changes over time, is an important process in climate change mitigation and adaptation. Existing approaches that use machine learning with Earth observation data for LCC rely on fully-annotated and segmented datasets. Creating these datasets requires a large amount of effort, and a lack of suitable datasets has become an obstacle in scaling the use of LCC. In this study, we propose Scene-to-Patch models: an alternative LCC approach utilising Multiple Instance Learning (MIL) that requires only high-level scene labels. This enables much faster development of new datasets whilst still providing segmentation through patch-level predictions, ultimately increasing the accessibility of using LCC for different scenarios. On the DeepGlobe-LCC dataset, our approach outperforms non-MIL baselines on both scene- and patch-level prediction. This work provides the foundation for expanding the use of LCC in climate change mitigation methods for technology, government, and academia.
Parnell, Katie; Fischer, Joel E; Clark, Jediah R; Bodenmann, Adrian; Trigo, Maria Jose Galvez; Brito, Mario; Soorati, Mohammad Divband; Plant, Katherine; Ramchurn, Sarvapali
Trustworthy UAV relationships: Applying the Schema Action World taxonomy to UAVs and UAV swarm operations Journal Article
In: International Journal of Human-Computer Interaction, 2022.
@article{soton468839,
title = {Trustworthy UAV relationships: Applying the Schema Action World taxonomy to UAVs and UAV swarm operations},
author = {Katie Parnell and Joel E Fischer and Jediah R Clark and Adrian Bodenmann and Maria Jose Galvez Trigo and Mario Brito and Mohammad Divband Soorati and Katherine Plant and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/468839/},
year = {2022},
date = {2022-07-01},
journal = {International Journal of Human-Computer Interaction},
abstract = {Human Factors play a significant role inthe development and integration of avionic systems to ensure that they are trusted and can be used effectively. As Unoccupied Aerial Vehicle (UAV) technology becomes increasingly important to the aviation domain this holds true. This study aims to gain an understanding of UAV operators?trust requirements when piloting UAVs by utilising a popular aviation interview methodology (Schema World Action Research Method), in combination with key questions on trust identified from the literature. Interviews were conducted with six UAVoperators, with a range of experience. This identified the importance of past experience to trust and the expectations that operators hold. Recommendations are made that target training to inform experience, in addition to the equipment, procedures and organisational standards that can aid in developing trustworthy systems. The methodology that was developed shows promise for capturing trust within human-automation interactions},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Human Factors play a significant role inthe development and integration of avionic systems to ensure that they are trusted and can be used effectively. As Unoccupied Aerial Vehicle (UAV) technology becomes increasingly important to the aviation domain this holds true. This study aims to gain an understanding of UAV operators?trust requirements when piloting UAVs by utilising a popular aviation interview methodology (Schema World Action Research Method), in combination with key questions on trust identified from the literature. Interviews were conducted with six UAVoperators, with a range of experience. This identified the importance of past experience to trust and the expectations that operators hold. Recommendations are made that target training to inform experience, in addition to the equipment, procedures and organisational standards that can aid in developing trustworthy systems. The methodology that was developed shows promise for capturing trust within human-automation interactions