Which Experiences Are Influential for Your Agent? Policy Iteration with Turn-over Dropout

TL;DR

This paper introduces PI+ToD, an efficient method for estimating the influence of experiences in reinforcement learning by combining policy iteration with turn-over dropout, demonstrated in MuJoCo environments.

Contribution

The paper proposes a novel approach, PI+ToD, that efficiently estimates experience influence in RL using turn-over dropout within policy iteration, reducing computational costs.

Findings

PI+ToD accurately estimates experience influence.

The method is computationally efficient in large experience buffers.

Effective in MuJoCo reinforcement learning environments.

Abstract

In reinforcement learning (RL) with experience replay, experiences stored in a replay buffer influence the RL agent's performance. Information about the influence is valuable for various purposes, including experience cleansing and analysis. One method for estimating the influence of individual experiences is agent comparison, but it is prohibitively expensive when there is a large number of experiences. In this paper, we present PI+ToD as a method for efficiently estimating the influence of experiences. PI+ToD is a policy iteration that efficiently estimates the influence of experiences by utilizing turn-over dropout. We demonstrate the efficiency of PI+ToD with experiments in MuJoCo environments.