Batch value function tournament for offline policy selection in reinforcement learning

Offline policy selection is a challenging open problem in reinforcement learning that has many important applications. The recently proposed Batch Value Function Tournament (BVFT) algorithm for batch learning offers some nice properties and can be applied to the model selection problem. In this thesis, we propose several changes to the original algorithm for adaptation to the task of offline model selection. We comprehensively experimented with the BVFT algorithm for policy selection across the various domains to evaluate and analyze the performance of BVFT. We show that BVFT achieves good performance in comparison with a number of state-of-the-art approaches. We demonstrate that BVFT is a reliable option to the problem of policy selection in offline reinforcement learning.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2021

Author dc:creator

• Zhang, Siyuan

Contributors dc:contributor

• Jiang, Nan

• Model Selection
• Reinforcement Learning, Batch, Offline, RL

Statement dc:rights

• Copyright 2021 Siyuan Zhang

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/110602

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/110602