Report Title: Towards Privacy-Preserving Dynamic Assortment Optimization under Inventory Constraints

Presenter: Yang Yu, Assistant Professor

Affiliation: City University of Hong Kong

Report Time: June 14, 2023 (Wednesday) at 9:00 AM

Report Location: Room 804, Science and Education Building A, Feicui Lake Campus

Report Abstract: Assortment optimization is a fundamental problem in the retailing industry, where an assortment is a set of (substitutable) items shown to a customer whose purchase behavior may be affected by the assortment. In real applications, customers’ choices/preferences over items are often not known a priori. Therefore, we need to learn the choice model of customers on the fly while making assortment decisions to maximize the revenue of the retailing platform. Moreover, when offering assortments, we often have to consider various practical issues such as limited resources and customers’ concerns about their privacy. Motivated by the demand from real retailing applications, in this talk, I will introduce our recent progress in dynamic assortment optimization under both inventory constraints and privacy protection constraints. Assuming the choice model is captured by a Multinomial Logit (MNL) model, our first attempt is to design a parallel MNL-bandit algorithm that offers assortments to a population of customers in parallel. By applying our algorithm, we can anonymize customers in the population while still learn the MNL model by some population statistics of customers’ choice decisions. We then adopt a more rigorous privacy mechanism, namely Differential Privacy, to design efficient online dual-mirror-descent algorithms for dynamic assortment optimization under inventory constraints. I will show that all our algorithms have provable regret upper bounds sublinear to the planning horizon $T$. Numerical simulations also demonstrate the effectiveness and efficiency of our algorithms. I will conclude this talk by discussing some interesting future directions to explore in the research of assortment optimization.

Personal Introduction: Dr. Yu Yang is currently an Assistant Professor with the School of Data Science at the City University of Hong Kong. His research interests lie in the algorithmic aspects of data mining and data science, especially in mining data of combinatorial structures and data-driven operations management. He has published his work in top data science venues such as ICML, NeurIPS, AISTATS, SIGMOD, VLDB, ICDE, KDD, TKDE, and TKDD. He obtained his Ph.D. in Computing Science from Simon Fraser University in Feb. 2019. Before that, he received his B.E. degree from Hefei University of Technology in 2010 and his M.E. degree from the University of Science and Technology of China in 2013, both in Computer Science.