This paper addresses the scalability challenge of architecture search by formulating the task in a differentiable manner. Unlike conventional approaches of applying evolution or reinforcement learning over a discrete and non-differentiable search space, our method is based on the continuous relaxation of the architecture representation, allowing efficient search of the architecture using gradient descent. Extensive experiments on CIFAR-10, ImageNet, Penn Treebank and WikiText-2 show that our algorithm excels in discovering high-performance convolutional architectures for image classification and recurrent architectures for language modeling, while being orders of magnitude faster than state-of-the-art non-differentiable techniques. Our implementation has been made publicly available to facilitate further research on efficient architecture search algorithms.
| Dataset | Unroll | Config | Download |
|---|---|---|---|
| Cifar10 | True | config | model | log |
| Dataset | Params(M) | Flops(G) | Top-1 Acc | Top-5 Acc | Subnet | Config | Download | Remarks |
|---|---|---|---|---|---|---|---|---|
| Cifar10 | 3.42 | 0.48 | 97.32 | 99.94 | mutable | config | model | log | MMRazor searched |
| Cifar10 | 3.83 | 0.55 | 97.27 | 99.98 | mutable | config | model | log | official |
@inproceedings{liu2018darts,
title={DARTS: Differentiable Architecture Search},
author={Liu, Hanxiao and Simonyan, Karen and Yang, Yiming},
booktitle={International Conference on Learning Representations},
year={2018}
}