English

L$^{2}$NAS: Learning to Optimize Neural Architectures via Continuous-Action Reinforcement Learning

Machine Learning 2021-09-28 v1 Computer Vision and Pattern Recognition

Abstract

Neural architecture search (NAS) has achieved remarkable results in deep neural network design. Differentiable architecture search converts the search over discrete architectures into a hyperparameter optimization problem which can be solved by gradient descent. However, questions have been raised regarding the effectiveness and generalizability of gradient methods for solving non-convex architecture hyperparameter optimization problems. In this paper, we propose L2^{2}NAS, which learns to intelligently optimize and update architecture hyperparameters via an actor neural network based on the distribution of high-performing architectures in the search history. We introduce a quantile-driven training procedure which efficiently trains L2^{2}NAS in an actor-critic framework via continuous-action reinforcement learning. Experiments show that L2^{2}NAS achieves state-of-the-art results on NAS-Bench-201 benchmark as well as DARTS search space and Once-for-All MobileNetV3 search space. We also show that search policies generated by L2^{2}NAS are generalizable and transferable across different training datasets with minimal fine-tuning.

Keywords

Cite

@article{arxiv.2109.12425,
  title  = {L$^{2}$NAS: Learning to Optimize Neural Architectures via Continuous-Action Reinforcement Learning},
  author = {Keith G. Mills and Fred X. Han and Mohammad Salameh and Seyed Saeed Changiz Rezaei and Linglong Kong and Wei Lu and Shuo Lian and Shangling Jui and Di Niu},
  journal= {arXiv preprint arXiv:2109.12425},
  year   = {2021}
}

Comments

Accepted as a Full Research Paper at CIKM 2021; 10 pages, 3 Figures, 5 Tables

R2 v1 2026-06-24T06:19:32.565Z