Rank-1 Approximation of Inverse Fisher for Natural Policy Gradients in Deep Reinforcement Learning

TL;DR

This paper introduces a scalable natural policy optimization method for deep reinforcement learning that uses a rank-1 approximation of the inverse Fisher Information Matrix, leading to faster convergence and better performance.

Contribution

The paper proposes a novel rank-1 approximation approach for inverse FIM in natural policy gradients, improving computational efficiency and convergence speed.

Findings

Achieves faster convergence than policy gradients.

Outperforms standard actor-critic and trust-region methods.

Maintains similar sample complexity as stochastic policy gradients.

Abstract

Natural gradients have long been studied in deep reinforcement learning due to their fast convergence properties and covariant weight updates. However, computing natural gradients requires inversion of the Fisher Information Matrix (FIM) at each iteration, which is computationally prohibitive in nature. In this paper, we present an efficient and scalable natural policy optimization technique that leverages a rank-1 approximation to full inverse-FIM. We theoretically show that under certain conditions, a rank-1 approximation to inverse-FIM converges faster than policy gradients and, under some conditions, enjoys the same sample complexity as stochastic policy gradient methods. We benchmark our method on a diverse set of environments and show that it achieves superior performance to standard actor-critic and trust-region baselines.