Efficient Entropy for Policy Gradient with Multidimensional Action Space

TL;DR

This paper introduces novel unbiased estimators for entropy in high-dimensional action spaces to improve exploration in policy gradient reinforcement learning, demonstrating enhanced performance with minimal extra computation.

Contribution

It develops new entropy estimators suitable for high-dimensional discrete action spaces and applies them to various policy models, improving exploration efficiency.

Findings

Entropy estimators significantly boost policy performance.

Methods are computationally efficient with marginal overhead.

Effective in multi-agent and complex environments.

Abstract

In recent years, deep reinforcement learning has been shown to be adept at solving sequential decision processes with high-dimensional state spaces such as in the Atari games. Many reinforcement learning problems, however, involve high-dimensional discrete action spaces as well as high-dimensional state spaces. This paper considers entropy bonus, which is used to encourage exploration in policy gradient. In the case of high-dimensional action spaces, calculating the entropy and its gradient requires enumerating all the actions in the action space and running forward and backpropagation for each action, which may be computationally infeasible. We develop several novel unbiased estimators for the entropy bonus and its gradient. We apply these estimators to several models for the parameterized policies, including Independent Sampling, CommNet, Autoregressive with Modified MDP, and Autoregressive with LSTM. Finally, we test our algorithms on two environments: a multi-hunter multi-rabbit grid game and a multi-agent multi-arm bandit problem. The results show that our entropy estimators substantially improve performance with marginal additional computational cost.