Towards Applicable Reinforcement Learning: Improving the Generalization and Sample Efficiency with Policy Ensemble

Arxiv: https://arxiv.org/abs/2205.09284

Code: https://github.com/microsoft/EPPO

Supplementary: PDF

Authors

• Zhengyu Yang (Shanghai Jiao Tong University) yzydestiny@sjtu.edu.cn
• Kan Ren (Microsoft Research) renkan@shanghaitech.edu.cn
• Xufang Luo (Microsoft Research) xufluo@microsoft.com
• Minghuan Liu (Shanghai Jiao Tong University) minghuanliu@sjtu.edu.cn
• Weiqing Liu (Microsoft Research) weiqing.liu@microsoft.com
• Jiang Bian (Microsoft Research) jiang.bian@microsoft.com
• Weinan Zhang (Shanghai Jiao Tong University) wnzhang@sjtu.edu.cn
• Dongsheng Li (Microsoft Research) dongsli@microsoft.com

Abstract

It is challenging for reinforcement learning (RL) algorithms to succeed in real-world applications like financial trading and logistic system due to the noisy observation and environment shifting between training and evaluation. Thus, it requires both high sample efficiency and generalization for resolving real-world tasks. However, directly applying typical RL algorithms can lead to poor performance in such scenarios. Considering the great performance of ensemble methods on both accuracy and generalization in supervised learning (SL), we design a robust and applicable method named Ensemble Proximal Policy Optimization (EPPO), which learns ensemble policies in an end-to-end manner. Notably, EPPO combines each policy and the policy ensemble organically and optimizes both simultaneously. In addition, EPPO adopts a diversity enhancement regularization over the policy space which helps to generalize to unseen states and promotes exploration. We theoretically prove EPPO increases exploration efficacy, and through comprehensive experimental evaluations on various tasks, we demonstrate that EPPO achieves higher efficiency and is robust for real-world applications compared with vanilla policy optimization algorithms and other ensemble methods. Code and supplemental materials are available at https://seqml.github.io/eppo.

Related Works

The Diversified Ensemble Neural Network
Proximal Policy Optimization Algorithms
Divide-and-Conquer Reinforcement Learning