【公开学术报告】Dimension Reduction in Contextual Online Learning via Nonparametric Variable Selection

Dimension Reduction in Contextual Online Learning via Nonparametric Variable Selection

Guest Speaker:   Wenhao LI (University of Toronto)

Date & Time:      9:30-11:00 (Beijing Time), Fri. 28th, Oct. 2022

Zoom Meeting: 879 956 01972（Password: 744631）

Click the Link:     https://us02web.zoom.us/j/87995601972

ABSTRACT

Contextual bandits has been a popular research topic in recent years. With the explosion of big data, the decision maker usually observes covariates before making decisions, which is helpful to assess the current situation and make better decisions. The current studies on contextual bandits can be categorized into linear bandits and nonparametric bandits. But there is a large gap between the assumptions of the two streams of literature. Our works fill the gap by considering the nonparametric functions with sparsity structures in covariates. To take advantage of the sparsity structure of the covariate, we propose a variable selection algorithm called BV-LASSO, which incorporates novel ideas such as binning and voting to apply LASSO (Least Absolute Shrinkage and Selection Operator) to nonparametric settings. We show the algorithm achieving a near-optimal regret. Our algorithm may serve as a general recipe to achieve dimension reduction via variable selection in nonparametric settings.

(This work is under revision in Journal of Machine Learning Research.)

Guest Bio:

Wenhao Li is currently a postdoctoral fellow at the Rotman School of Management, University of Toronto. He received his bachelor’s degree in Automation from Zhejiang University and Ph.D. in Management Sciences from the City University of Hong Kong. His research mainly focuses on data-driven decision making and its application in healthcare. For more information, please see his personal website, https://wenhaoli666.github.io/.