12/4/25 | 4:15pm | E51-145
Michael Lingzhi Li
Assistant Professor of Business Administration
Harvard Business School
Abstract: This study addresses facility location problems in the context of drone delivery systems, which involves multiple stations and demand locations. Drones are deployed at selected stations to serve demand locations, with service time varying based on the distance between stations and demand locations. Each demand location has two attributes: demand level (indicating required service instances) and time-dependent service success probability (p-value). The p-values are heterogeneous across locations and time intervals, often initially unknown or uncertain, which exist in the objective of maximizing the sum of expected successful services across all agents. To handle the uncertainty in p-values, the problem is formulated as a combinatorial multi-armed bandit (CMAB) problem. Here, each agent servicing a demand location at a specific time represents a base arm, while the collective allocation strategy forms a super arm. In this paper, we develop a two-stage stochastic programming scheme to iteratively explore p-values and exploit optimal locations. To deal with the curse of dimensionality, we propose a novel grouping-based estimation algorithm that enhances computational efficiency while keeping the optimality gap below 5% generally. Experiments in out-of-hospital cardiac arrests (OHCA) demonstrate that this approach consistently outperforms baseline methods in balancing exploration-exploitation trade-offs and minimizing cumulative regrets. The framework effectively addresses dynamic environments where success probabilities require adaptive estimation through strategic sampling and allocation adjustments.
Bio: Michael Lingzhi Li is an Assistant Professor in the Technology and Operations Management unit at Harvard Business School. His research focuses on the end-to-end development of decision algorithms based on machine learning, causal inference and operations research. He examines the implementation of such algorithms in high-stakes decision-making, with a focus on healthcare applications. He is the recipient of awards including the Edelman Laureate, the Pierskalla Award, and the Innovative Applications in Analytics Award.