We study the benefits of different sparse architectures for deep reinforcement learning. In particular, we focus on image-based domains where spatially-biased and fully-connected architectures are common. Using these and several other architectures of equal capacity, we show that sparse structure has a significant effect on learning performance. We also observe that choosing the best sparse architecture for a given domain depends on whether the hidden layer weights are fixed or learned.
@article{arxiv.2501.16729,
title = {On the Interplay Between Sparsity and Training in Deep Reinforcement Learning},
author = {Fatima Davelouis and John D. Martin and Michael Bowling},
journal= {arXiv preprint arXiv:2501.16729},
year = {2025}
}