Location:Olin 305
When: March 7th at 1:30 p.m.
Title: Offline Data-Driven Decision-Making: Some Challenges and Solutions
Abstract: Offline data-driven decision-making is concerned with finding a policy that maximizes the total utility using a pre-collected dataset. A fundamental challenge behind this task is the so-called distributional mismatch, stemming from two primary sources: the limited exploration of the offline data distribution and unmeasured confounding. In this talk, I will use assortment optimization and reinforcement learning as examples to illustrate these issues and present our proposals to address them. If time permits, I will also discuss the phenomenon called “blessing from Human-AI interaction” in offline data-driven decision-making under unmeasured confounding and introduce the framework of super decision-learning.
Zoom link: https://wse.zoom.us/j/94601022340