In this paper, we investigate the fundamental question: To what extent are gradient-based neural architecture search (NAS) techniques applicable to RL? Using the original DARTS as a convenient baseline, we discover that the discrete architectures found can achieve up to 250% performance compared to manual architecture designs on both discrete and continuous action space environments across off-policy and on-policy RL algorithms, at only 3x more computation time. Furthermore, through numerous ablation studies, we systematically verify that not only does DARTS correctly upweight operations during its supernet phrase, but also gradually improves resulting discrete cells up to 30x more efficiently than random search, suggesting DARTS is surprisingly an effective tool for improving architectures in RL.
@article{arxiv.2106.02229,
title = {Differentiable Architecture Search for Reinforcement Learning},
author = {Yingjie Miao and Xingyou Song and John D. Co-Reyes and Daiyi Peng and Summer Yue and Eugene Brevdo and Aleksandra Faust},
journal= {arXiv preprint arXiv:2106.02229},
year = {2022}
}
Comments
Published as a conference paper at the first Automated Machine Learning Conference (AutoML-Conf) 2022. Code can be found at https://github.com/google/brain_autorl/tree/main/rl_darts