A Variance-Reduced Cubic-Regularized Newton for Policy Optimization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Sun, Cheng, Zhang, Zhen, Yang, Shaofu
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915388599566336
author Sun, Cheng
Zhang, Zhen
Yang, Shaofu
author_facet Sun, Cheng
Zhang, Zhen
Yang, Shaofu
contents In this paper, we study a second-order approach to policy optimization in reinforcement learning. Existing second-order methods often suffer from suboptimal sample complexity or rely on unrealistic assumptions about importance sampling. To overcome these limitations, we propose VR-CR-PN, a variance-reduced cubic-regularized policy Newton algorithm. To the best of our knowledge, this is the first algorithm that integrates Hessian-aided variance reduction with second-order policy optimization, effectively addressing the distribution shift problem and achieving best-known sample complexity under general nonconvex conditions but without the need for importance sampling. We theoretically establish that VR-CR-PN achieves a sample complexity of $\tilde{\mathcal{O}}(ε^{-3})$ to reach an $ε$-second-order stationary point, significantly improving upon the previous best result of $\tilde{\mathcal{O}}(ε^{-3.5})$ under comparable assumptions. As an additional contribution, we introduce a novel Hessian estimator for the expected return function, which admits a uniform upper bound independent of the horizon length $H$, allowing the algorithm to achieve horizon-independent sample complexity.
format Preprint
id arxiv_https___arxiv_org_abs_2507_10120
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle A Variance-Reduced Cubic-Regularized Newton for Policy Optimization
Sun, Cheng
Zhang, Zhen
Yang, Shaofu
Machine Learning
Artificial Intelligence
In this paper, we study a second-order approach to policy optimization in reinforcement learning. Existing second-order methods often suffer from suboptimal sample complexity or rely on unrealistic assumptions about importance sampling. To overcome these limitations, we propose VR-CR-PN, a variance-reduced cubic-regularized policy Newton algorithm. To the best of our knowledge, this is the first algorithm that integrates Hessian-aided variance reduction with second-order policy optimization, effectively addressing the distribution shift problem and achieving best-known sample complexity under general nonconvex conditions but without the need for importance sampling. We theoretically establish that VR-CR-PN achieves a sample complexity of $\tilde{\mathcal{O}}(ε^{-3})$ to reach an $ε$-second-order stationary point, significantly improving upon the previous best result of $\tilde{\mathcal{O}}(ε^{-3.5})$ under comparable assumptions. As an additional contribution, we introduce a novel Hessian estimator for the expected return function, which admits a uniform upper bound independent of the horizon length $H$, allowing the algorithm to achieve horizon-independent sample complexity.
title A Variance-Reduced Cubic-Regularized Newton for Policy Optimization
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2507.10120