Online Program Home
My Program

Abstract Details

Activity Number: 517 - Deep Learning: Advances and Applications
Type: Contributed
Date/Time: Wednesday, July 31, 2019 : 10:30 AM to 12:20 PM
Sponsor: Section on Statistical Learning and Data Science
Abstract #305038
Title: Online Batch Decision Making with High-Dimensional Covariates
Author(s): Chi-Hua Wang* and Guang Cheng
Companies: Purdue University and Purdue Statistics
Keywords: multi-armed bandits with covariates; statistical decision-making; high-dimensional statistics; LASSO; batches

Motivated by online advertising and personalized medicine, this paper proposes an algorithm for the batched bandit problem with high dimensional covariate, named as {\em Teamwork LASSO Bandit}. Specifically, we consider a multi-armed bandit problem in which arms are sampled in batches instead of one at a time. Meanwhile, high-dimensional covariate of each user is available to online decision makers. In this setup, the objective is to learn a teamwork strategy, together with a batch sampling scheme, that minimizes the accumulated total regret. To address a batch version of explore-exploit dilemma, we propose a novel framework to perform batch sampling by carefully alternating teamwork stage and greedy stage during the whole sequential decision making process. In theory, under a fundamental teamwork strategy, an asymptotic upper bound is provided for the expected cumulative total regret of the resulting algorithm.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program