Dynamic Sparse Training for Deep Reinforcement Learning

TL;DR

This paper introduces a dynamic sparse training method for deep reinforcement learning that accelerates training, reduces resource consumption, and improves performance by adapting network topology during training.

Contribution

It is the first to apply dynamic sparse training to DRL, enabling faster training and better efficiency compared to dense networks.

Findings

Achieves higher performance than dense methods on continuous control tasks.

Reduces parameter count and FLOPs by 50%.

Speeds up learning, reaching dense agent performance with 40-50% fewer training steps.

Abstract

Deep reinforcement learning (DRL) agents are trained through trial-and-error interactions with the environment. This leads to a long training time for dense neural networks to achieve good performance. Hence, prohibitive computation and memory resources are consumed. Recently, learning efficient DRL agents has received increasing attention. Yet, current methods focus on accelerating inference time. In this paper, we introduce for the first time a dynamic sparse training approach for deep reinforcement learning to accelerate the training process. The proposed approach trains a sparse neural network from scratch and dynamically adapts its topology to the changing data distribution during training. Experiments on continuous control tasks show that our dynamic sparse agents achieve higher performance than the equivalent dense methods, reduce the parameter count and floating-point operations (FLOPs) by 50%, and have a faster learning speed that enables reaching the performance of dense agents with 40-50% reduction in the training steps.