Nearly Horizon-Free Offline Reinforcement Learning

Authors: Tongzheng Ren, Jialian Li, Bo Dai, Simon S. Du, Sujay Sanghavi

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical For offline policy evaluation, we obtain an O error bound for the plug-in estimator, which matches the lower bound up to logarithmic factors and does not have additional dependency on poly (H, S, A, d) in higher-order term. For offline policy optimization, we obtain an O suboptimality gap for the empirical optimal policy, which approaches the lower bound up to logarithmic factors and a high-order term, improving upon the best known result by [1] that has additional poly (H, S, d) factors in the main term. To the best of our knowledge, these are the first set of nearly horizon-free bounds for episodic time-homogeneous offline tabular MDP and linear MDP with anchor points.
Researcher Affiliation Collaboration Tongzheng Ren1 Jialian Li2 Bo Dai3 Simon S. Du4 Sujay Sanghavi1, 5 1 UT Austin 2 Tsinghua University 3 Google Research, Brain Team 4 University of Washington 5 Amazon Search
Pseudocode No The paper does not include any pseudocode or clearly labeled algorithm blocks.
Open Source Code No The paper does not include any statement about releasing source code or provide any links to a code repository.
Open Datasets No The paper is theoretical and focuses on sample complexity bounds and theoretical guarantees. It discusses "collected K episodes data" as input for offline reinforcement learning, but it does not specify a publicly available or open dataset used for empirical training of a model, nor does it provide access information for such a dataset.
Dataset Splits No The paper is theoretical and does not describe experiments that would require dataset splits. Therefore, it does not provide information about training, validation, or test dataset splits.
Hardware Specification No The paper is theoretical and does not describe empirical experiments, thus it does not provide any hardware specifications used for running experiments.
Software Dependencies No The paper is theoretical and does not describe empirical experiments. Therefore, it does not list any specific software dependencies with version numbers.
Experiment Setup No The paper is theoretical and does not describe empirical experiments. Therefore, it does not provide specific experimental setup details, hyperparameters, or training configurations.