configs | ||
data | ||
exps-cnn | ||
exps-rnn | ||
lib | ||
paddlepaddle | ||
scripts-cluster | ||
scripts-cnn | ||
scripts-rnn | ||
.gitignore | ||
LICENSE | ||
README.md |
Searching for A Robust Neural Architecture in Four GPU Hours
We propose A Gradient-based neural architecture search approach using Differentiable Architecture Sampler (GDAS). Please find details in our paper.

Requirements
- PyTorch 1.0.1
- Python 3.6
- opencv
conda install pytorch torchvision cuda100 -c pytorch
Usages
Train the searched CNN on CIFAR
CUDA_VISIBLE_DEVICES=0 bash ./scripts-cnn/train-cifar.sh GDAS_FG cifar10 cut
CUDA_VISIBLE_DEVICES=0 bash ./scripts-cnn/train-cifar.sh GDAS_F1 cifar10 cut
CUDA_VISIBLE_DEVICES=0 bash ./scripts-cnn/train-cifar.sh GDAS_V1 cifar100 cut
Train the searched CNN on ImageNet
CUDA_VISIBLE_DEVICES=0,1,2,3 bash ./scripts-cnn/train-imagenet.sh GDAS_F1 52 14 B128 -1
CUDA_VISIBLE_DEVICES=0,1,2,3 bash ./scripts-cnn/train-imagenet.sh GDAS_V1 50 14 B256 -1
Evaluate a trained CNN model
CUDA_VISIBLE_DEVICES=0 python ./exps-cnn/evaluate.py --data_path $TORCH_HOME/cifar.python --checkpoint ${checkpoint-path}
CUDA_VISIBLE_DEVICES=0 python ./exps-cnn/evaluate.py --data_path $TORCH_HOME/ILSVRC2012 --checkpoint ${checkpoint-path}
CUDA_VISIBLE_DEVICES=0 python ./exps-cnn/evaluate.py --data_path $TORCH_HOME/ILSVRC2012 --checkpoint GDAS-V1-C50-N14-ImageNet.pth
Train the searched RNN
CUDA_VISIBLE_DEVICES=0 bash ./scripts-rnn/train-PTB.sh DARTS_V1
CUDA_VISIBLE_DEVICES=0 bash ./scripts-rnn/train-PTB.sh DARTS_V2
CUDA_VISIBLE_DEVICES=0 bash ./scripts-rnn/train-PTB.sh GDAS
CUDA_VISIBLE_DEVICES=0 bash ./scripts-rnn/train-WT2.sh DARTS_V1
CUDA_VISIBLE_DEVICES=0 bash ./scripts-rnn/train-WT2.sh DARTS_V2
CUDA_VISIBLE_DEVICES=0 bash ./scripts-rnn/train-WT2.sh GDAS
Training Logs
You can find some training logs in ./data/logs/
.
You can also find some pre-trained models in Google Driver.
Experimental Results

Citation
If you find that this project (GDAS) helps your research, please cite the paper:
@inproceedings{dong2019search,
title={Searching for A Robust Neural Architecture in Four GPU Hours},
author={Dong, Xuanyi and Yang, Yi},
booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
pages={1761--1770},
year={2019}
}