Algorithmic Fairness in Temporal Resource Allocation

Ashwin Kumar

Abstract: There has been a significant body of research on improving social welfare in resource allocation, but much of it has focused on single-shot allocation scenarios, where a given pool of resources must be divided equitably. In contrast, my research aims to address the unique challenges posed by temporal resource allocation problems that involve many repeated allocations, with both resources and beneficiaries able to re-enter the market at different points in time. Automated algorithms are often employed to guide resource allocation in these scenarios by estimating and comparing utilities of different allocations, making algorithmic fairness a concern as well. In this work, I aim to improve long-term social welfare in addition to maximizing the utility of such systems through the lens of pre-, in-, and post-processing fairness. I propose a simple incentive-based approach for post-processing fairness with black-box value functions, outperforming existing baselines in a ridesharing application. I discuss two other research thrusts using fairness-aware dataset balancing for pre-processing fairness and learning non-myopic fairness policies for in-processing fairness. Combining all of these approaches, my goal is to present a holistic view of improving social welfare in temporal resource allocation through the lens of algorithmic fairness.