Abstract
Motivated by applications in social services and beyond, we study data-driven policies for the online allocation of scarce resources—such as housing, wireless bandwidth, or cloud computing resources—to heterogeneous individuals or agents under uncertainty. Leveraging observational data collected in deployment, we develop a unified framework that integrates predictive modeling and optimization to learn allocation policies that maximize long-run outcomes subject to capacity and fairness constraints. Our approach constructs simple, interpretable waitlist-based policies that balance individual benefit with the opportunity cost of limited resources and achieve strong asymptotic optimality guarantees.
A key challenge in these systems is that distribution shifts arise from e.g., evolving measurement processes or changes in user behavior, altering both covariate distributions and their relationship with outcomes between training and deployment. We address this using a distributionally robust optimization approach that hedges against such shifts while allowing practitioners to tune trade-offs between efficiency and robustness.
Empirical results on real-world homelessness data demonstrate significant improvements in allocation outcomes and robustness with minimal cost of fairness. We further discuss how the proposed framework extends to engineering domains, including network resource allocation in communication systems, or scheduling in cloud computing platforms.
Bio
Phebe Vayanos is an associate professor of industrial & systems engineering and computer science at the University of Southern California where she holds a Viterbi Early Career Chair in Engineering. She is also a co-director of CAIS, the Center for Artificial Intelligence in Society and a co-director of the ORAI Program in Operations Research and Artificial Intelligence at USC. Her research is focused on optimization and its interface with machine learning, causal inference, and economics. Prior to joining USC, she was lecturer in the Operations Research and Statistics Group at the MIT Sloan School of Management, and a postdoctoral research associate in the Operations Research Center at MIT. She holds a master’s degree in electrical and electronic engineering and a doctoral degree in Computing, both from Imperial College London. She is a TED AI speaker, a recipient of the NSF CAREER award, and the Imperial College Emerging Alumni Leader Award, among others. She is an associate editor for Management Science, Operations Research, Operations Research Letters, and Computational Management Science.
