Efficient Neural Architecture Search via Parameter Sharing

TL;DR

ENAS introduces a fast, cost-effective neural architecture search method using parameter sharing, achieving state-of-the-art results on language modeling and competitive performance on image classification with significantly reduced computational resources.

Contribution

The paper presents ENAS, a novel neural architecture search approach that drastically reduces search time and cost through parameter sharing, outperforming or matching existing methods.

Findings

Achieves a test perplexity of 55.8 on Penn Treebank.

Attains a test error of 2.89% on CIFAR-10.

Reduces computational cost by 1000x compared to standard NAS.

Abstract

We propose Efficient Neural Architecture Search (ENAS), a fast and inexpensive approach for automatic model design. In ENAS, a controller learns to discover neural network architectures by searching for an optimal subgraph within a large computational graph. The controller is trained with policy gradient to select a subgraph that maximizes the expected reward on the validation set. Meanwhile the model corresponding to the selected subgraph is trained to minimize a canonical cross entropy loss. Thanks to parameter sharing between child models, ENAS is fast: it delivers strong empirical performances using much fewer GPU-hours than all existing automatic model design approaches, and notably, 1000x less expensive than standard Neural Architecture Search. On the Penn Treebank dataset, ENAS discovers a novel architecture that achieves a test perplexity of 55.8, establishing a new state-of-the-art among all methods without post-training processing. On the CIFAR-10 dataset, ENAS designs novel architectures that achieve a test error of 2.89%, which is on par with NASNet (Zoph et al., 2018), whose test error is 2.65%.