ResearchTrend.AI
  • Papers
  • Communities
  • Events
  • Blog
  • Pricing
Papers
Communities
Social Events
Terms and Conditions
Pricing
Parameter LabParameter LabTwitterGitHubLinkedInBlueskyYoutube

© 2025 ResearchTrend.AI, All rights reserved.

  1. Home
  2. Papers
  3. 2105.14016
89
20
v1v2v3 (latest)

Sample-Efficient Reinforcement Learning for Linearly-Parameterized MDPs with a Generative Model

28 May 2021
Bingyan Wang
Yuling Yan
Jianqing Fan
ArXiv (abs)PDFHTML
Abstract

The curse of dimensionality is a widely known issue in reinforcement learning (RL). In the tabular setting where the state space S\mathcal{S}S and the action space A\mathcal{A}A are both finite, to obtain a nearly optimal policy with sampling access to a generative model, the minimax optimal sample complexity scales linearly with ∣S∣×∣A∣|\mathcal{S}|\times|\mathcal{A}|∣S∣×∣A∣, which can be prohibitively large when S\mathcal{S}S or A\mathcal{A}A is large. This paper considers a Markov decision process (MDP) that admits a set of state-action features, which can linearly express (or approximate) its probability transition kernel. We show that a model-based approach (resp. ~ Q-learning) provably learns an ε\varepsilonε-optimal policy (resp. ~ Q-function) with high probability as soon as the sample size exceeds the order of K(1−γ)3ε2\frac{K}{(1-\gamma)^{3}\varepsilon^{2}}(1−γ)3ε2K​ (resp. ~ K(1−γ)4ε2\frac{K}{(1-\gamma)^{4}\varepsilon^{2}}(1−γ)4ε2K​), up to some logarithmic factor. Here KKK is the feature dimension and γ∈(0,1)\gamma\in(0,1)γ∈(0,1) is the discount factor of the MDP. Both sample complexity bounds are provably tight, and our result for the model-based approach matches the minimax lower bound. Our results show that for arbitrarily large-scale MDP, both the model-based approach and Q-learning are sample-efficient when KKK is relatively small, and hence the title of this paper.

View on arXiv
Comments on this paper