Neural Network Compression for Reinforcement Learning Tasks
This work addresses efficiency needs in RL applications like robotics, but it is incremental as it applies standard pruning techniques to RL.
The paper tackled the problem of reducing neural network size for reinforcement learning tasks to improve latency and energy efficiency, achieving up to a 400-fold reduction in network size.
In real applications of Reinforcement Learning (RL), such as robotics, low latency and energy efficient inference is very desired. The use of sparsity and pruning for optimizing Neural Network inference, and particularly to improve energy and latency efficiency, is a standard technique. In this work, we perform a systematic investigation of applying these optimization techniques for different RL algorithms in different RL environments, yielding up to a 400-fold reduction in the size of neural networks.