Efficient Neural Architecture Search via Proximal Iterations

Requirements

Python >= 3.5.5, PyTorch == 0.3.1, torchvision == 0.2.0

Dataset

Prepare dataset firstly, dataset links can be seen in our Appendix.

Neural Architecture Search

python train_search.py

Architecture Evaluation

python train.py --auxiliary --cutout # CIFAR-10 python train_tinyimagenet.py --auxiliary # Tiny ImageNet

Architecture Test

python test.py --auxiliary --model_path cifar10_model.pt # CIFAR-10 python test_tinyimagenet.py --auxiliary --model_path imagenet_model.pt # Tiny Imagenet

Reference

@techreport{yao2019differentiable,
  title     = {Efficient Neural Architecture Search via Proximal Iterations},
  author    = {Yao, Q. and Xu, J. and Tu, W.-W. and Zhu, Z.},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year      = {2020}
}

@TechReport{quanming2018auto,
  author      = {Yao , Q. and Wang, M.},
  title       = {Taking human out of learning applications: A survey on automated machine learning},
  institution = {Arxiv: 1810.13306},
  year        = {2018},
}

Acknowledgement

The codes of this paper are based on the codes of DARTS (https://github.com/quark0/darts). We appreciate DARTS's codes and thank the authors of DARTS.