mmrazor/configs/nas/mmcls/darts

DARTS

DARTS: Differentiable Architecture Search

Abstract

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.

Get Started

Step 1: Supernet training on Cifar-10

CUDA_VISIBLE_DEVICES=0,1,2,3 PORT=29500 ./tools/dist_train.sh \
  configs/nas/mmcls/darts/darts_supernet_unroll_1xb96_cifar10.py 4 \
  --work-dir $WORK_DIR

Step 2: Subnet retraining on Cifar-10

CUDA_VISIBLE_DEVICES=0,1,2,3 PORT=29500 ./tools/dist_train.sh \
  configs/nas/mmcls/darts/darts_subnet_1xb96_cifar10_2.0.py 4 \
  --work-dir $WORK_DIR \
  --cfg-options model.init_cfg.checkpoint=$STEP2_CKPT

Step 3: Subnet inference on Cifar-10

CUDA_VISIBLE_DEVICES=0 PORT=29500 ./tools/dist_test.sh \
  configs/nas/mmcls/darts/darts_subnet_1xb96_cifar10_2.0.py \
  none 1 --work-dir $WORK_DIR \
  --cfg-options model.init_cfg.checkpoint=$STEP2_CKPT

Results and models

Supernet

Dataset	Unroll	Config	Download
Cifar10	True	config	model | log

Subnet

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

Citation

@inproceedings{liu2018darts,
  title={DARTS: Differentiable Architecture Search},
  author={Liu, Hanxiao and Simonyan, Karen and Yang, Yiming},
  booktitle={International Conference on Learning Representations},
  year={2018}
}