Bellman Unbiasedness: Toward Provably Efficient Distributional Reinforcement Learning with General Value Function Approximation

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Cho, Taehyun, Han, Seungyub, Ju, Seokhun, Kim, Dohyeong, Lee, Kyungjae, Lee, Jungwoo
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866916733611147264
author Cho, Taehyun
Han, Seungyub
Ju, Seokhun
Kim, Dohyeong
Lee, Kyungjae
Lee, Jungwoo
author_facet Cho, Taehyun
Han, Seungyub
Ju, Seokhun
Kim, Dohyeong
Lee, Kyungjae
Lee, Jungwoo
contents Distributional reinforcement learning improves performance by capturing environmental stochasticity, but a comprehensive theoretical understanding of its effectiveness remains elusive. In addition, the intractable element of the infinite dimensionality of distributions has been overlooked. In this paper, we present a regret analysis of distributional reinforcement learning with general value function approximation in a finite episodic Markov decision process setting. We first introduce a key notion of $\textit{Bellman unbiasedness}$ which is essential for exactly learnable and provably efficient distributional updates in an online manner. Among all types of statistical functionals for representing infinite-dimensional return distributions, our theoretical results demonstrate that only moment functionals can exactly capture the statistical information. Secondly, we propose a provably efficient algorithm, $\texttt{SF-LSVI}$, that achieves a tight regret bound of $\tilde{O}(d_E H^{\frac{3}{2}}\sqrt{K})$ where $H$ is the horizon, $K$ is the number of episodes, and $d_E$ is the eluder dimension of a function class.
format Preprint
id arxiv_https___arxiv_org_abs_2407_21260
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Bellman Unbiasedness: Toward Provably Efficient Distributional Reinforcement Learning with General Value Function Approximation
Cho, Taehyun
Han, Seungyub
Ju, Seokhun
Kim, Dohyeong
Lee, Kyungjae
Lee, Jungwoo
Machine Learning
Artificial Intelligence
Distributional reinforcement learning improves performance by capturing environmental stochasticity, but a comprehensive theoretical understanding of its effectiveness remains elusive. In addition, the intractable element of the infinite dimensionality of distributions has been overlooked. In this paper, we present a regret analysis of distributional reinforcement learning with general value function approximation in a finite episodic Markov decision process setting. We first introduce a key notion of $\textit{Bellman unbiasedness}$ which is essential for exactly learnable and provably efficient distributional updates in an online manner. Among all types of statistical functionals for representing infinite-dimensional return distributions, our theoretical results demonstrate that only moment functionals can exactly capture the statistical information. Secondly, we propose a provably efficient algorithm, $\texttt{SF-LSVI}$, that achieves a tight regret bound of $\tilde{O}(d_E H^{\frac{3}{2}}\sqrt{K})$ where $H$ is the horizon, $K$ is the number of episodes, and $d_E$ is the eluder dimension of a function class.
title Bellman Unbiasedness: Toward Provably Efficient Distributional Reinforcement Learning with General Value Function Approximation
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2407.21260