English

Spectral Decomposition Representation for Reinforcement Learning

Machine Learning 2023-03-08 v2 Machine Learning

Abstract

Representation learning often plays a critical role in reinforcement learning by managing the curse of dimensionality. A representative class of algorithms exploits a spectral decomposition of the stochastic transition dynamics to construct representations that enjoy strong theoretical properties in an idealized setting. However, current spectral methods suffer from limited applicability because they are constructed for state-only aggregation and derived from a policy-dependent transition kernel, without considering the issue of exploration. To address these issues, we propose an alternative spectral method, Spectral Decomposition Representation (SPEDER), that extracts a state-action abstraction from the dynamics without inducing spurious dependence on the data collection policy, while also balancing the exploration-versus-exploitation trade-off during learning. A theoretical analysis establishes the sample efficiency of the proposed algorithm in both the online and offline settings. In addition, an experimental investigation demonstrates superior performance over current state-of-the-art algorithms across several benchmarks.

Keywords

Cite

@article{arxiv.2208.09515,
  title  = {Spectral Decomposition Representation for Reinforcement Learning},
  author = {Tongzheng Ren and Tianjun Zhang and Lisa Lee and Joseph E. Gonzalez and Dale Schuurmans and Bo Dai},
  journal= {arXiv preprint arXiv:2208.09515},
  year   = {2023}
}

Comments

ICLR 2023. The first two authors contribute equally

R2 v1 2026-06-25T01:49:50.206Z