Department of Industrial Engineering & Decision Analytics [Joint IEDA/ISOM seminar] - Online Algorithms for Correlated Arrivals

10:30am - 11:30am
Room 5510 (lift 25-26)

Online algorithms are ubiquitous in e-commerce systems, where inventory, fulfillment, and assortment decisions must be committed to in real-time without knowing future demand. For many of these problems, there is a lack of decision-making approaches that are aware of correlations in demand over time. This is partly due to the challenge of model selection, but even given an explicit model of correlated demand, it is unclear how online decisions should be prescribed.

We present two new approaches that overcome these challenges by implicitly capturing the correlations in demand arrivals over time. First, we present a non-parametric framework that models only the distribution of the total arrivals of each "type", and then by making assumptions on how these arrivals are interleaved, leads to correlation-aware online decisions. Second, we discuss a model-free "reinforcement learning" approach that directly optimizes for average performance over a collection of past (correlated) arrival sequences, while restricting to a carefully curated class of policies. We discuss the benefits of each approach, both of which leverage the ability to evaluate counterfactual policy performance in inventory and fulfillment problems. Finally, we discuss details of the first approach specifically for the online matching (fulfillment) problem, which is based on joint work with Ali Aouad (London Business School -> MIT).

讲者/ 表演者:
Prof. Will Ma
Columbia Business School, Decision, Risk, and Operations

Will Ma is an Associate Professor of Decision, Risk, and Operations at Columbia Business School. His research centers around online algorithms in e-commerce systems, both for supply-side problems like inventory and fulfillment, and revenue management problems like dynamic assortment optimization. He specializes in designing simple online algorithms with performance guarantees, that can be tuned to historical data. Will also has miscellaneous experience as a professional poker player, video-game startup founder, and karaoke bar pianist.

语言
英文
适合对象
教职员
研究生
主办单位
Department of Industrial Engineering & Decision Analytics
信息,商业统计及营运学系
新增活动
请各校内团体将活动发布至大学活动日历。