Action-depedent Control Variates for Policy Optimization via Stein's Identity

Abstract

Policy gradient methods have achieved remarkable successes in solving challenging reinforcement learning problems. However, it still often suffers from the large variance issue on policy gradient estimation, which leads to poor sample efficiency during training. In this work, we propose a control variate method to effectively reduce variance for policy gradient methods. Motivated by the Stein's identity, our method extends the previous control variate methods used in REINFORCE and advantage actor-critic by introducing more general action-dependent baseline functions. Empirical studies show that our method significantly improves the sample efficiency of the state-of-the-art policy gradient approaches.

arXiv commentThe first two authors contributed equally. Author ordering determined by coin flip over a Google Hangout. Accepted by ICLR 2018

Cited by in corpus (10)

• A Survey and Critique of Multiagent Deep Reinforcement Learning
• Beyond variance reduction: Understanding the true impact of baselines on policy optimization
• P3O: Policy-on Policy-off Policy Optimization
• Variance Reduction for Evolution Strategies via Structured Control Variates
• Trajectory-wise Control Variates for Variance Reduction in Policy Gradient Methods
• Improving Text Generation with Student-Forcing Optimal Transport
• Scalable Approximate Inference and Some Applications
• Discrete Action On-Policy Learning with Action-Value Critic
• Unifying Gradient Estimators for Meta-Reinforcement Learning via Off-Policy Evaluation
• Co-Adaptation of Algorithmic and Implementational Innovations in Inference-based Deep Reinforcement Learning