Publications

@article{soton467975,

title = {From Intelligent Agents to Trustworthy Human-Centred Multiagent Systems},

author = {Mohammad Divband Soorati and Enrico Gerding and Enrico Marchioni and Pavel Naumov and Timothy Norman and Sarvapali Ramchurn and Baharak Rastegari and Adam Sobey and Sebastian Stein and Danesh Tarapore and Vahid Yazdanpanah and Jie Zhang},

url = {https://eprints.soton.ac.uk/467975/},

year  = {2022},

date = {2022-07-01},

journal = {AI Communications},

abstract = {The Agents, Interaction and Complexity research group at the University of Southampton has a long track record of research in multiagent systems (MAS). We have made substantial scientific contributions across learning in MAS, game-theoretic techniques for coordinating agent systems, and formal methods for representation and reasoning. We highlight key results achieved by the group and elaborate on recent work and open research challenges in developing trustworthy autonomous systems and deploying human-centred AI systems that aim to support societal good.},

keywords = {Artificial Intelligence, Distributed Artificial Intelligence, Intelligent Agents, Multiagent Systems, Trustworthy Autonomous Systems},

pubstate = {published},

tppubtype = {article}

}

@article{soton458023,

title = {Non-markovian reward modelling from trajectory labels via interpretable multiple instance learning},

author = {Joseph Early and Tom Bewley and Christine Evers and Sarvapali Ramchurn},

url = {https://eprints.soton.ac.uk/458023/},

year  = {2022},

date = {2022-05-01},

journal = {arXiv},

abstract = {We generalise the problem of reward modelling (RM) for reinforcement learning (RL) to handle non-Markovian rewards. Existing work assumes that human evaluators observe each step in a trajectory independently when providing feedback on agent behaviour. In this work, we remove this assumption, extending RM to include hidden state information that captures temporal dependencies in human assessment of trajectories. We then show how RM can be approached as a multiple instance learning (MIL) problem, and develop new MIL models that are able to capture the time dependencies in labelled trajectories. We demonstrate on a range of RL tasks that our novel MIL models can reconstruct reward functions to a high level of accuracy, and that they provide interpretable learnt hidden information that can be used to train high-performing agent policies.},

note = {20 pages (9 main content; 2 references; 9 appendix). 11 figures (8 main content; 3 appendix)},

keywords = {cs.AI, cs.LG},

pubstate = {published},

tppubtype = {article}

}

@article{soton455806,

title = {Mechanism design for efficient offline and online allocation of electric vehicles to charging stations},

author = {Emmanouil S Rigas and Enrico H Gerding and Sebastian Stein and Sarvapali D Ramchurn and Nick Bassiliades},

url = {https://eprints.soton.ac.uk/455806/},

year  = {2022},

date = {2022-03-01},

journal = {Energies},

volume = {15},

number = {5},

abstract = {ensuremath<pensuremath>The industry related to electric vehicles (EVs) has seen a substantial increase in recent years, as such vehicles have the ability to significantly reduce total COensuremath<subensuremath>2ensuremath</subensuremath> emissions and the related global warming effect. In this paper, we focus on the problem of allocating EVs to charging stations, scheduling and pricing their charging. Specifically, we developed a Mixed Integer Program (MIP) which executes offline and optimally allocates EVs to charging stations. On top, we propose two alternative mechanisms to price the electricity the EVs charge. The first mechanism is a typical fixed-price one, while the second is a variation of the Vickrey?Clark?Groves (VCG) mechanism. We also developed online solutions that incrementally call the MIP-based algorithm and solve it for branches of EVs. In all cases, the EVs? aim is to minimize the price to pay and the impact on their driving schedule, acting as self-interested agents. We conducted a thorough empirical evaluation of our mechanisms and we observed that they had satisfactory scalability. Additionally, the VCG mechanism achieved an up to 2.2% improvement in terms of the number of vehicles that were charged compared to the fixed-price one and, in cases where the stations were congested, it calculated higher prices for the EVs and provided a higher profit for the stations, but lower utility to the EVs. However, in a theoretical evaluation, we proved that the variant of the VCG mechanism being proposed in this paper still guaranteed truthful reporting of the EVs? preferences. In contrast, the fixed-price one was found to be vulnerable to agents? strategic behavior as non-truthful EVs can charge instead of truthful ones. Finally, we observed the online algorithms to be, on average, at 95.6% of the offline ones in terms of the average number of serviced EVs.ensuremath</pensuremath>},

note = {Funding Information: 

Funding: This research study was co-financed by Greece and the European Union (European Social Fund–ESF) through the Operational Programme ?Human Resources Development, Education and Lifelong Learning? in the context of the project ?Reinforcement of Postdoctoral Researchers-2nd Cycle? (MIS-5033021), implemented by State Scholarships Foundation (IKY). 

Copyright 2022 Elsevier B.V., All rights reserved.},

keywords = {Charging, Electric Vehicles, Fixed price, mechanism design, Scheduling, VCG},

pubstate = {published},

tppubtype = {article}

}

@article{soton450597,

title = {Trustworthy human-AI partnerships},

author = {Sarvapali Ramchurn and Sebastian Stein and Nicholas R Jennings},

url = {https://eprints.soton.ac.uk/450597/},

year  = {2021},

date = {2021-08-01},

journal = {iScience},

volume = {24},

number = {8},

abstract = {In this paper, we foreground some of the key research challenges that arise in the design of trustworthy human-AI partnerships. In particular, we focus on the challenges in designing human-AI partnerships that need to be addressed to help humans and organisations trust their machine counterparts individually or as a collective (e.g., as robot teams or groups of software agents). We also aim to identify the risks associated with human-AI partnerships and therefore determine the associated measures to mitigate these risks. By so doing, we will trigger new avenues of research that will address the key barriers to the adoption of AI-based systems more widely in our daily lives and in industry.},

keywords = {Artificial Intelligence, Human-Computer Interaction, Sociology},

pubstate = {published},

tppubtype = {article}

}

@article{soton449675,

title = {Resilient intersection management with multi-vehicle collision avoidance},

author = {Phuriwat Worrawichaipat and Enrico Gerding and Ioannis Kaparias and Sarvapali Ramchurn},

url = {https://eprints.soton.ac.uk/449675/},

year  = {2021},

date = {2021-06-01},

journal = {Frontiers in Sustainable Cities},

volume = {3},

abstract = {In this paper, we propose a novel decentralised agent-based mechanism for road intersection management for connected autonomous vehicles. In our work we focus on road obstructions causing major traffic delays. In doing so, we propose the first decentralised mechanism able to maximise the overall vehicle throughput at intersections in the presence of obstructions. The distributed algorithm transfers most of the computational cost from the intersection manager to the driving agents, thereby improving scalability. Our realistic empirical experiments using SUMO show that, when an obstacle is located at the entrance or in the middle of the intersection, existing state of the art algorithms and traffic lights show a reduced throughput of 65?90% from the optimal point without obstructions while our mechanism can maintain the throughput upensuremath<br/ensuremath>Q7 to 94?99%.},

keywords = {Computer science, intersection management, multi-agent systems, simulation experiments, Transportation},

pubstate = {published},

tppubtype = {article}

}

@article{soton452050,

title = {Large-scale, dynamic and distributed coalition formation with spatial and†temporal constraints},

author = {Luca Capezzuto and Danesh Tarapore and Sarvapali Ramchurn},

url = {https://eprints.soton.ac.uk/452050/},

year  = {2021},

date = {2021-05-01},

journal = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)},

pages = {108–125},

abstract = {The†Coalition Formation with Spatial and Temporal constraints Problem†(CFSTP) is a multi-agent task allocation problem in which few agents have to perform many tasks, each with its deadline and workload. To maximize the number of completed tasks, the agents need to cooperate by forming, disbanding and reforming coalitions. The original mathematical programming formulation of the CFSTP is difficult to implement, since it is lengthy and based on the problematic Big-M method. In this paper, we propose a compact and easy-to-implement formulation. Moreover, we design D-CTS, a distributed version of the state-of-the-art CFSTP algorithm. Using public London Fire Brigade records, we create a dataset with 347588 tasks and a test framework that simulates the mobilization of firefighters in dynamic environments. In problems with up†to 150 agents and 3000 tasks, compared to DSA-SDP, a state-of-the-art distributed algorithm, D-CTS completes†3.79%$pm$[42.22%,1.96%]†more tasks, and is one order of magnitude more efficient in terms of communication overhead and time complexity. D-CTS sets the first large-scale, dynamic and distributed CFSTP benchmark.ensuremath<br/ensuremath>},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{soton467373,

title = {Anytime and efficient multi-agent coordination for disaster response},

author = {Luca Capezzuto and Danesh Tarapore and Sarvapali D Ramchurn},

url = {https://eprints.soton.ac.uk/467373/},

year  = {2021},

date = {2021-03-01},

journal = {SN Computer Science},

volume = {2},

number = {3},

abstract = {The Coalition Formation with Spatial and Temporal constraints Problem (CFSTP) is a multi-agent task allocation problem where the tasks are spatially distributed, with deadlines and workloads, and the number of agents is typically much smaller than the number of tasks. To maximise the number of completed tasks, the agents may have to schedule coalitions. The state-of-the-art CFSTP solver, the Coalition Formation with Look-Ahead (CFLA) algorithm, has two main limitations. First, its time complexity is exponential with the number of agents. Second, as we show, its look-ahead technique is not effective in real-world scenarios, such as open multi-agent systems, where new tasks can appear at any time. In this work, we study its design and define a variant, called Coalition Formation with Improved Look-Ahead (CFLA2), which achieves better performance. Since we cannot eliminate the limitations of CFLA in CFLA2, we also develop a novel algorithm to solve the CFSTP, the first to be simultaneously anytime, efficient and with convergence guarantee, called Cluster-based Task Scheduling (CTS). In tests where the look-ahead technique is highly effective, CTS completes up to 30% (resp. 10%) more tasks than CFLA (resp. CFLA2) while being up to 4 orders of magnitude faster. We also propose S-CTS, a simplified but parallel variant of CTS with even lower time complexity. Using scenarios generated by the RoboCup Rescue Simulation, we show that S-CTS is at most 10% less performing than high-performance algorithms such as Binary Max-Sum and DSA, but up to 2 orders of magnitude faster. Our results affirm CTS as the new state-of-the-art algorithm to solve the CFSTP.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{soton445733,

title = {Partner selection in self-organised wireless sensor networks for opportunistic energy negotiation: A multi-armed bandit based approach},

author = {Andre P Ortega and Sarvapali Ramchurn and Long Tran-Thanh and Geoff Merrett},

url = {https://eprints.soton.ac.uk/445733/},

year  = {2021},

date = {2021-03-01},

journal = {Ad Hoc Networks},

volume = {112},

abstract = {The proliferation of ?Things? over a network creates the Internet of Things (IoT), where sensors integrate to collect data from the environment over long periods of time. The growth of IoT applications will inevitably involve co-locating multiple wireless sensor networks, each serving different applications with, possibly, different needs and constraints. Since energy is scarce in sensor nodes equipped with non-rechargeable batteries, energy harvesting technologies have been the focus of research in recent years. However, new problems arise as a result of their wide spatio-temporal variation. Such a shortcoming can be avoided if co-located networks cooperate with each other and share their available energy. Due to their unique characteristics and different owners, recently, we proposed a negotiation approach to deal with conflict of preferences. Unfortunately, negotiation can be impractical with a large number of participants, especially in an open environment. Given this, we introduce a new partner selection technique based on multi-armed bandits (MAB), that enables each node to learn the strategy that optimises its energy resources in the long term. Our results show that the proposed solution allows networks to repeatedly learn the current best energy partner in a dynamic environment. The performance of such a technique is evaluated through simulation and shows that a network can achieve an efficiency of 72% against the optimal strategy in the most challenging scenario studied in this work.},

keywords = {Agent-based sensor network, Automated negotiation, Energy management, Multi-armed bandit based learning, Reinforcement Learning, Wireless sensor networks},

pubstate = {published},

tppubtype = {article}

}

@article{soton445995,

title = {Optimising daily fantasy sports teams with artificial intelligence},

author = {Ryan James Beal and Timothy Norman and Sarvapali Ramchurn},

url = {https://eprints.soton.ac.uk/445995/},

year  = {2020},

date = {2020-12-01},

journal = {International Journal of Computer Science in Sport},

volume = {19},

number = {2},

abstract = {This paper outlines a novel approach to optimising teams for Daily Fantasy Sports (DFS) contests. To this end, we propose a number of new models and algorithms to solve the team formation problems posed by DFS. Specifically, we focus on the National Football League (NFL) and predict the performance of real-world players to form the optimal fantasy team using mixed-integer programming. We test our solutions using real-world data-sets from across four seasons (2014-2017). We highlight the advantage that can be gained from using our machine-based methods and show that our solutions outperform existing benchmarks, turning a profit in up to 81.3% of DFS game-weeks over a season},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{soton446078,

title = {A critical comparison of machine learning classifiers to predict match outcomes in the NFL},

author = {Ryan James Beal and Timothy Norman and Sarvapali Ramchurn},

url = {https://eprints.soton.ac.uk/446078/},

year  = {2020},

date = {2020-12-01},

journal = {International Journal of Computer Science in Sport},

volume = {19},

number = {2},

abstract = {In this paper, we critically evaluate the performance of nine machine learning classification techniques when applied to the match outcome prediction problem presented by American Football. Specifically, we implement and test nine techniques using real-world datasets of 1280 games over 5 seasons from the National Football League (NFL). We test the nine different classifier techniques using a total of 42 features for each team and we find that the best performing algorithms are able to improve one previous published works. The algoriothms achieve an accuracy of between 44.64% for a Guassian Process classifier to 67.53% with a Na"ive Bayes classifer. We also test each classifier on a year by year basis and compare our results to those of the bookmakers and other leading academic papers.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{oluwasuji2020solving,

title = {Solving the fair electric load shedding problem in developing countries},

author = {Olabambo Ifeoluwa Oluwasuji and Obaid Malik and Jie Zhang and Sarvapali Dyanand Ramchurn},

year  = {2020},

date = {2020-01-01},

journal = {Autonomous Agents and Multi-Agent Systems},

volume = {34},

number = {1},

pages = {12},

publisher = {Springer},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{DBLP:journals/aamas/OluwasujiMZR20,

title = {Solving the fair electric load shedding problem in developing countries},

author = {Olabambo Ifeoluwa Oluwasuji and Obaid Malik and Jie Zhang and Sarvapali Dyanand Ramchurn},

url = {https://doi.org/10.1007/s10458-019-09428-8},

doi = {10.1007/s10458-019-09428-8},

year  = {2020},

date = {2020-01-01},

journal = {Auton. Agents Multi Agent Syst.},

volume = {34},

number = {1},

pages = {12},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{DBLP:journals/simpra/SeitaridisRBR20,

title = {An agent-based negotiation scheme for the distribution of electric 

 vehicles across a set of charging stations},

author = {Andreas Seitaridis and Emmanouil S Rigas and Nick Bassiliades and Sarvapali D Ramchurn},

url = {https://doi.org/10.1016/j.simpat.2019.102040},

doi = {10.1016/j.simpat.2019.102040},

year  = {2020},

date = {2020-01-01},

journal = {Simul. Model. Pract. Theory},

volume = {100},

pages = {102040},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{DBLP:journals/tits/KoufakisRBR20,

title = {Offline and Online Electric Vehicle Charging Scheduling With V2V Energy Transfer},

author = {Alexandros Koufakis and Emmanouil S Rigas and Nick Bassiliades and Sarvapali D Ramchurn},

url = {https://doi.org/10.1109/TITS.2019.2914087},

doi = {10.1109/TITS.2019.2914087},

year  = {2020},

date = {2020-01-01},

journal = {IEEE Trans. Intell. Transp. Syst.},

volume = {21},

number = {5},

pages = {2128–2138},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{DBLP:journals/corr/abs-2003-10294,

title = {Optimising Game Tactics for Football},

author = {Ryan Beal and Georgios Chalkiadakis and Timothy J Norman and Sarvapali D Ramchurn},

url = {https://arxiv.org/abs/2003.10294},

year  = {2020},

date = {2020-01-01},

journal = {CoRR},

volume = {abs/2003.10294},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{DBLP:journals/corr/abs-2003-13806,

title = {Anytime and Efficient Coalition Formation with Spatial and Temporal 

 Constraints},

author = {Luca Capezzuto and Danesh Tarapore and Sarvapali D Ramchurn},

url = {https://arxiv.org/abs/2003.13806},

year  = {2020},

date = {2020-01-01},

journal = {CoRR},

volume = {abs/2003.13806},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{DBLP:journals/corr/abs-2007-09715,

title = {Mechanism Design for Efficient Online and Offline Allocation of Electric 

 Vehicles to Charging Stations},

author = {Emmanouil Rigas and Enrico Gerding and Sebastian Stein and Sarvapali D Ramchurn and Nick Bassiliades},

url = {https://arxiv.org/abs/2007.09715},

year  = {2020},

date = {2020-01-01},

journal = {CoRR},

volume = {abs/2007.09715},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{beal2019artificial,

title = {Artificial intelligence for team sports: a survey},

author = {Ryan Beal and Timothy J Norman and Sarvapali D Ramchurn},

year  = {2019},

date = {2019-01-01},

journal = {The Knowledge Engineering Review},

volume = {34},

publisher = {Cambridge University Press},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{koufakis2019offline,

title = {Offline and Online Electric Vehicle Charging Scheduling With V2V Energy Transfer},

author = {Alexandros-Michail Koufakis and Emmanouil S Rigas and Nick Bassiliades and Sarvapali D Ramchurn},

year  = {2019},

date = {2019-01-01},

journal = {IEEE Transactions on Intelligent Transportation Systems},

publisher = {IEEE},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{seitaridis2019agent,

title = {An Agent-based Negotiation Scheme for the Distribution of Electric Vehicles Across a Set of Charging Stations},

author = {Andreas Seitaridis and Emmanouil S Rigas and Nick Bassiliades and Sarvapali D Ramchurn},

year  = {2019},

date = {2019-01-01},

journal = {Simulation Modelling Practice and Theory},

pages = {102040},

publisher = {Elsevier},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{abioye2019effects,

title = {Effects of Varying Noise Levels and Lighting Levels on Multimodal Speech and Visual Gesture Interaction with Aerobots},

author = {Ayodeji Opeyemi Abioye and Stephen D Prior and Peter Saddington and Sarvapali D Ramchurn},

year  = {2019},

date = {2019-01-01},

journal = {Applied Sciences},

volume = {9},

number = {10},

pages = {2066},

publisher = {Multidisciplinary Digital Publishing Institute},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{soton422097,

title = {Algorithms for electric vehicle scheduling in large-scale mobility-on-demand schemes},

author = {Emmanouil Rigas and Sarvapali Ramchurn and Nick Bassiliades},

url = {https://eprints.soton.ac.uk/422097/},

year  = {2018},

date = {2018-09-01},

journal = {Artificial Intelligence},

volume = {262},

pages = {248–278},

abstract = {We study a setting where Electric Vehicles (EVs) can be hired to drive from pick-up to drop-off points in a Mobility-on-Demand (MoD) scheme. The goal of the system is, either to maximize the number of customers that are serviced, or the total EV utilization. To do so, we characterise the optimisation problem as a max-flow problem in order to determine the set of feasible trips given the available EVs at each location. We then model and solve the EV-to-trip scheduling problem offline and optimally using Mixed Integer Programming (MIP) techniques and show that the solution scales up to medium sized problems. Given this, we develop two non-optimal algorithms, namely an incremental-MIP algorithm for medium to large problems and a greedy heuristic algorithm for very large problems. Moreover, we develop a tabu search-based local search technique to further improve upon and compare against the solution of the non-optimal algorithms. We study the performance of these algorithms in settings where either battery swap or battery charge at each station is used to cope with the EVs' limited driving range. Moreover, in settings where EVs need to be scheduled online, we propose a novel algorithm that accounts for the uncertainty in future trip requests. All algorithms are empirically evaluated using real-world data of locations of shared vehicle pick-up and drop-off stations. In our experiments, we observe that when all EVs carry the same battery which is large enough for the longest trips, the greedy algorithm with battery swap with the max-flow solution as a pre-processing step, provides the optimal solution. At the same time, the greedy algorithm with battery charge is close to the optimal (97% on average) and is further improved when local search is used. When some EVs do not have a large enough battery to execute some of the longest trips, the incremental-MIP generates solutions slightly better than the greedy, while the optimal algorithm is the best but scales up to medium sized problems only. Moreover, the online algorithm is shown to be on average at least 90% of the optimal. Finally, the greedy algorithm scales to 10-times more tasks than the incremental-MIP and 1000-times more than the static MIP in reasonable time.},

keywords = {Electric Vehicles, Heuristics, Mobility on Demand, optimisation, Scheduling},

pubstate = {published},

tppubtype = {article}

}

@article{soton422967,

title = {Seeing (movement) is believing: the effect of motion on perception of automatic systems performance},

author = {Pedro Garcia Garcia and Enrico Costanza and Jhim Verame and Diana Nowacka and Sarvapali D Ramchurn},

url = {https://eprints.soton.ac.uk/422967/},

year  = {2018},

date = {2018-04-01},

journal = {Human-Computer Interaction},

pages = {1-51},

abstract = {ensuremath<pensuremath>In this article, we report on one lab study and seven follow-up studies on a crowdsourcing platform designed to investigate the potential of animation cues to influence users? perception of two smart systems: a handwriting recognition and a part-of-speech tagging system. Results from the first three studies indicate that animation cues can influence a participant?s perception of both systems? performance. The subsequent three studies, designed to try and identify an explanation for this effect, suggest that this effect is related to the participants? mental model of the smart system. The last two studies were designed to characterize the effect more in detail, and they revealed that different amounts of animation do not seem to create substantial differences and that the effect persists even when the system?s performance decreases, but only when the difference in performance level between the systems being compared is small.ensuremath</pensuremath>},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{BICEGO2018317b,

title = {On the distinctiveness of the electricity load profile},

author = {M. Bicego and A. Farinelli and E. Grosso and D. Paolini and S. D. Ramchurn},

url = {http://www.sciencedirect.com/science/article/pii/S0031320317303904},

doi = {https://doi.org/10.1016/j.patcog.2017.09.039},

issn = {0031-3203},

year  = {2018},

date = {2018-01-01},

journal = {Pattern Recognition},

volume = {74},

number = {Supplement C},

pages = {317–325},

keywords = {Biometrics, Classification, Energy market, Load profile, Pre-processing},

pubstate = {published},

tppubtype = {article}

}

@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}

}

@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}

}

@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}

}

@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}

}

@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}

}

@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}

}