Introduction
Allocation of scare resources is prevalent in our lives.
From allocating drivers and vans for delivery of goods, to the allocation of beds and doctors in hospitals, to the allocation of paramedics and ambulances in disaster response. Take Emergency Services as an example. They play a critical role by dispatching vehicles and qualified professionals to emergency situations. While providing skilled and well-equipped staff in a short time is key in most scenarios, they operate with limited resources and hence it is crucial to ensure that right level of resources is used to deal with a situation, while maintaining the resilience of the system for upcoming emergencies.
An efficient allocation of resources makes the best use of the limited resources available and consequently enhances the system’s sustainability. In many applications, it is important for the users to perceive the allocation as ‘’fair’’ for the allocation mechanism to be deployed and sustained successfully. Efficiency and fairness have been studied extensively. However, most studies make assumptions that do not hold in many real-life scenarios. An allocation is resilient if, should a problem occur (e.g. an ambulance breaks down and becomes unavailable), it can be amended with minimal loss to efficiency and fairness. Very few studies address resilience in limited settings.
We aim to address the existing limitations of literature by considering settings that are dynamic and incorporate uncertainty.
Dynamic: The set of resources and tasks/agents (e.g. ambulances and patients) change over time. Allocation of resources are temporal (e.g. an ambulance is allocated to a patient for a length of time and will be available again after the patient reaches the hospital). Resources may have spatial attributes and their location plays a role in how an allocation should be made and how to evaluate the quality of an allocation.
Partial and dynamic preferences: Agents’ preferences over available resources, and resources’ priorities over agents, may not be known to the system designer or even to the agents and resources themselves. There could be privacy or security concerns which raise the need to design systems that can produce good results without access to full preferences. Preferences can change over time when more information becomes available or the status of the agent changes. For example, a patient in an emergency may be happy to go to any hospital with an available bed, but when the situation stabilises, they prefer to stay in a hospital closer to home.
Research Goals
The main goal of the project is to use multi-agent systems and machine learning techniques to design fair, efficient and resilient allocation of scarce, and possibly heavily constrained, resources in dynamic settings with partial preferences. Our research objective could include (1) formalising what constitutes a resilient mobility/logistics system, (2) analysing the trade-off between resilience and efficiency and fairness, (3) investigating the resilient, efficient and fair allocation of scare resources, (4) utilising agent-based simulations to evaluate various coordination and resource allocation mechanisms or policies. The specifics of the project can be adapted to the PhD candidate's skills and interests.
References:
References
Akbarpour, Li and Oveis Gharan, "Thickness and information in dynamic matching markets", J. Political Econ, 2019.
Garg and Murhekar: "On fair and efficient allocations of indivisible goods", AAAI, 2021
Zeng and Psomas, "Fairness-Efficiency Tradeoffs in Dynamic Fair Division", EC, 2020
Lodi, Olivier, et al. "Fairness over time in dynamic resource allocation with an application in healthcare". arXiv, 2022
Yang, R., Ford, B. J., Tambe, M., & Lemieux, A. (2014, May). Adaptive resource allocation for wildlife protection against illegal poachers. In AAAMAS (pp. 453-460).