Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Xu, Wanqiao, Dong, Shi, Van Roy, Benjamin
Format: Preprint
Veröffentlicht: 2022
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2211.15931
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866908589823623168
author Xu, Wanqiao
Dong, Shi
Van Roy, Benjamin
author_facet Xu, Wanqiao
Dong, Shi
Van Roy, Benjamin
contents We develop an extension of posterior sampling for reinforcement learning (PSRL) that is suited for a continuing agent-environment interface and integrates naturally into agent designs that scale to complex environments. The approach, continuing PSRL, maintains a statistically plausible model of the environment and follows a policy that maximizes expected $γ$-discounted return in that model. At each time, with probability $1-γ$, the model is replaced by a sample from the posterior distribution over environments. For a choice of discount factor that suitably depends on the horizon $T$, we establish an $\tilde{O}(τS \sqrt{A T})$ bound on the Bayesian regret, where $S$ is the number of environment states, $A$ is the number of actions, and $τ$ denotes the reward averaging time, which is a bound on the duration required to accurately estimate the average reward of any policy. Our work is the first to formalize and rigorously analyze the resampling approach with randomized exploration.
format Preprint
id arxiv_https___arxiv_org_abs_2211_15931
institution arXiv
publishDate 2022
record_format arxiv
spellingShingle Posterior Sampling for Continuing Environments
Xu, Wanqiao
Dong, Shi
Van Roy, Benjamin
Machine Learning
We develop an extension of posterior sampling for reinforcement learning (PSRL) that is suited for a continuing agent-environment interface and integrates naturally into agent designs that scale to complex environments. The approach, continuing PSRL, maintains a statistically plausible model of the environment and follows a policy that maximizes expected $γ$-discounted return in that model. At each time, with probability $1-γ$, the model is replaced by a sample from the posterior distribution over environments. For a choice of discount factor that suitably depends on the horizon $T$, we establish an $\tilde{O}(τS \sqrt{A T})$ bound on the Bayesian regret, where $S$ is the number of environment states, $A$ is the number of actions, and $τ$ denotes the reward averaging time, which is a bound on the duration required to accurately estimate the average reward of any policy. Our work is the first to formalize and rigorously analyze the resampling approach with randomized exploration.
title Posterior Sampling for Continuing Environments
topic Machine Learning
url https://arxiv.org/abs/2211.15931