GST: Group-Sparse Training for Accelerating Deep Reinforcement Learning
arXiv:2101.09650
Abstract
Deep reinforcement learning (DRL) has shown remarkable success in sequential decision-making problems but suffers from a long training time to obtain such good performance. Many parallel and distributed DRL training approaches have been proposed to solve this problem, but it is difficult to utilize them on resource-limited devices. In order to accelerate DRL in real-world edge devices, memory bandwidth bottlenecks due to large weight transactions have to be resolved. However, previous iterative pruning not only shows a low compression ratio at the beginning of training but also makes DRL training unstable. To overcome these shortcomings, we propose a novel weight compression method for DRL training acceleration, named group-sparse training (GST). GST selectively utilizes block-circulant compression to maintain a high weight compression ratio during all iterations of DRL training and dynamically adapt target sparsity through reward-aware pruning for stable training. Thanks to the features, GST achieves a 25 \%p 41.5 \%p higher average compression ratio than the iterative pruning method without reward drop in Mujoco Halfcheetah-v2 and Mujoco humanoid-v2 environment with TD3 training.
10 pages, 10 figures, CVPR 2021 Submitted
References in corpus (6)
- Challenges of Real-World Reinforcement Learning
- Sample-Efficient Reinforcement Learning with Stochastic Ensemble Value Expansion
- Structured Transforms for Small-Footprint Deep Learning
- PoPS: Policy Pruning and Shrinking for Deep Reinforcement Learning
- The Ingredients of Real-World Robotic Reinforcement Learning
- ACDC: A Structured Efficient Linear Layer