Learning Near Optimal Policies with Low Inherent Bellman Error

Authors: Andrea Zanette, Alessandro Lazaric, Mykel Kochenderfer, Emma Brunskill

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical We study the exploration problem with approximate linear action-value functions in episodic reinforcement learning under the notion of low inherent Bellman error... While computational tractability questions remain open for the MDP setting, this enriches the class of MDPs with a linear representation for the action-value function where statistically efficient reinforcement learning is possible.
Researcher Affiliation Collaboration 1Stanford University 2Facebook Artificial Intelligence Research. Correspondence to: Andrea Zanette <zanette@stanford.edu>.
Pseudocode Yes Algorithm 1 ELEANOR
Open Source Code No Although ELEANOR is proved to be near optimal, it is difficult to implement the algorithm efficiently. This should not be seen as a fundamental barrier, however... For now, we leave this to future work.
Open Datasets No The paper is theoretical and does not report on experiments with datasets, thus no information on training datasets is provided.
Dataset Splits No The paper is theoretical and does not report on experiments with datasets, thus no information on training/validation/test splits is provided.
Hardware Specification No The paper is theoretical and does not report on experiments or provide any hardware specifications.
Software Dependencies No The paper is theoretical and does not report on experiments or provide any specific software dependencies with version numbers.
Experiment Setup No The paper is theoretical and does not report on experiments or provide any experimental setup details such as hyperparameters.