Batch size-invariance for policy optimization

TL;DR

This paper introduces a method to make policy optimization algorithms like PPO batch size-invariant by decoupling the policy update control from the behavior policy, enabling more efficient use of data.

Contribution

The authors propose a novel approach to achieve batch size-invariance in policy optimization algorithms by decoupling the proximal and behavior policies.

Findings

Algorithms can be made batch size-invariant with the proposed method.

Decoupling policies improves efficiency in using stale data.

Experimental results support the effectiveness of the approach.

Abstract

We say an algorithm is batch size-invariant if changes to the batch size can largely be compensated for by changes to other hyperparameters. Stochastic gradient descent is well-known to have this property at small batch sizes, via the learning rate. However, some policy optimization algorithms (such as PPO) do not have this property, because of how they control the size of policy updates. In this work we show how to make these algorithms batch size-invariant. Our key insight is to decouple the proximal policy (used for controlling policy updates) from the behavior policy (used for off-policy corrections). Our experiments help explain why these algorithms work, and additionally show how they can make more efficient use of stale data.