Dr. Yaakov Malinovsky and his collaborators, Dr. Goldenshluger Alexander (University of Haifa) and Dr. Zeevi Assaf (Columbia Univiversity), are the recipients of a four-year grant from the US-Israel Binational Science Foundation.
Project Title: Minimax online learning policies in stochastic sequential selection and assignment
Abstract: This research proposal focuses on a novel set of problems in stochastic sequential decision making that require “real time” learning of unknown probabilistic modeling primitives. In particular, we consider problems of stochastic sequential selection and assignment under incomplete information and so-called “bandit feedback.” These problems arise in many applied contexts, ranging from sequential auctions (that can be modeled as optimal stopping problems), organ transplant exchanges (that can be modeled as stochastic sequential assignment problems) through online marketplaces such as TaskRabbit.com (as well as numerous other platforms in the labor market that have emerged in recent years). When full information exists, such classes of online decision making problems are typically solved using dynamic programming principles. Far less is known in the case of incomplete information, especially when one considers solving said problems in an online manner. The purpose of this project is to shed some light on the learning theoretic aspects and to provide an analytical framework that supports structural insights, elucidates the learning challenges that are salient to these problems, and develop solution concepts that are efficient and computationally tractable.