English

On Using Hamiltonian Monte Carlo Sampling for Reinforcement Learning Problems in High-dimension

Machine Learning 2022-03-29 v3 Systems and Control Systems and Control Optimization and Control

Abstract

Value function based reinforcement learning (RL) algorithms, for example, QQ-learning, learn optimal policies from datasets of actions, rewards, and state transitions. However, when the underlying state transition dynamics are stochastic and evolve on a high-dimensional space, generating independent and identically distributed (IID) data samples for creating these datasets poses a significant challenge due to the intractability of the associated normalizing integral. In these scenarios, Hamiltonian Monte Carlo (HMC) sampling offers a computationally tractable way to generate data for training RL algorithms. In this paper, we introduce a framework, called \textit{Hamiltonian QQ-Learning}, that demonstrates, both theoretically and empirically, that QQ values can be learned from a dataset generated by HMC samples of actions, rewards, and state transitions. Furthermore, to exploit the underlying low-rank structure of the QQ function, Hamiltonian QQ-Learning uses a matrix completion algorithm for reconstructing the updated QQ function from QQ value updates over a much smaller subset of state-action pairs. Thus, by providing an efficient way to apply QQ-learning in stochastic, high-dimensional settings, the proposed approach broadens the scope of RL algorithms for real-world applications.

Keywords

Cite

@article{arxiv.2011.05927,
  title  = {On Using Hamiltonian Monte Carlo Sampling for Reinforcement Learning Problems in High-dimension},
  author = {Udari Madhushani and Biswadip Dey and Naomi Ehrich Leonard and Amit Chakraborty},
  journal= {arXiv preprint arXiv:2011.05927},
  year   = {2022}
}
R2 v1 2026-06-23T20:06:02.764Z