学术动态

3月23日 | 周文卓 Estimating Optimal Infinite Horizon Dynamic Treatment Regimes via pT-Learning

时间:2022.3.23 20:00-21:00

地点:腾讯会议 209 322 588

题目:Estimating OptimalInfinite Horizon Dynamic Treatment Regimes via pT-Learning

报告人:周文卓 伊利诺伊大学香槟分校 博士生

摘要:Recent advances in mobilehealth (mHealth) technology provide an effective way to monitor individuals'health statuses and deliver just-in-time personalized interventions. However,the practical use of mHealth technology raises unique challenges to existingmethodologies on learning an optimal dynamic treatment regime. Many mHealthapplications involve decision-making with large numbers of intervention optionsand under an infinite time horizon setting where the number of decision stagesdiverges to infinity. In addition, temporary medication shortages may causeoptimal treatments to be unavailable, while it is unclear what alternatives canbe used. To address these challenges, we propose a Proximal  Temporal consistency Learning (pT-Learning)framework to estimate an optimal regime that is adaptively adjusted betweendeterministic and stochastic sparse policy models. The resulting minimaxestimator avoids the double sampling issue in the existing algorithms. It canbe further simplified and can easily incorporate off-policy data withoutmismatched distribution corrections. We study theoretical properties of thesparse policy and establish finite-sample bounds on the excess risk andperformance error. The proposed method is implemented by our proximalDTRpackage and is evaluated through extensive simulation studies and the OhioT1DMmHealth dataset. This is a joint work with Prof. Ruoqing Zhu and Prof. AnnieQu. 

报告人简介:Wenzhuo Zhou is a finalyear Ph.D. student in the Department of Statistics at University of IllinoisUrbana Champaign. Prior to that, he obtained the bachelor’s degrees inMathematics and Economics from Southwestern University of Finance andEconomics. His research mainly focuses on the dynamic treatment regimes,reinforcement learning, and recommender system. His works have been published in some statistical journals, includingJournal of the American Statistical Association, Biometrika and The R journal,etc.

发布者:张瑛发布时间:2022-03-08浏览次数:10