SNAS: Stochastic Neural Architecture Search

ICLR 2019  ·  Sirui Xie, Hehui Zheng, Chunxiao Liu, Liang Lin ·

We propose Stochastic Neural Architecture Search (SNAS), an economical end-to-end solution to Neural Architecture Search (NAS) that trains neural operation parameters and architecture distribution parameters in same round of back-propagation, while maintaining the completeness and differentiability of the NAS pipeline. In this work, NAS is reformulated as an optimization problem on parameters of a joint distribution for the search space in a cell. To leverage the gradient information in generic differentiable loss for architecture search, a novel search gradient is proposed. We prove that this search gradient optimizes the same objective as reinforcement-learning-based NAS, but assigns credits to structural decisions more efficiently. This credit assignment is further augmented with locally decomposable reward to enforce a resource-efficient constraint. In experiments on CIFAR-10, SNAS takes less epochs to find a cell architecture with state-of-the-art accuracy than non-differentiable evolution-based and reinforcement-learning-based NAS, which is also transferable to ImageNet. It is also shown that child networks of SNAS can maintain the validation accuracy in searching, with which attention-based NAS requires parameter retraining to compete, exhibiting potentials to stride towards efficient NAS on big datasets. We have released our implementation at https://github.com/SNAS-Series/SNAS-Series.

PDF Abstract ICLR 2019 PDF ICLR 2019 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Neural Architecture Search NAS-Bench-201, ImageNet-16-120 SNAS Accuracy (Test) 43.16 # 29

Results from Other Papers


Task Dataset Model Metric Name Metric Value Rank Source Paper Compare
Neural Architecture Search NAS-Bench-201, CIFAR-10 SNAS Accuracy (Test) 92.77 # 25
Accuracy (Val) 90.10 # 19
Neural Architecture Search NAS-Bench-201, CIFAR-100 SNAS Accuracy (Test) 69.34 # 28
Accuracy (Val) 69.69 # 24

Methods