GenNAS-Zero

This repository is the official implementation of Generic Neural Architecture Search via Regression (NeurIPS'21 spotlight | Openreview | Arxiv version). Besides, a faster method is introduced Arxiv version.

Todo

• NASBench-101
• Training
• requirement.sh
• Exploration
• NASBench-201
• NDS
• TransNASBench-Micro
• NASBench-MB
• Preprocessing data
• ReadMe
• Standalone Instruction

Standalone Instruction

Please run notebooks/standalone.ipynb

Start a Simple Zero Cost Proxy Sarch

bash requirement.sh
conda activate gennaszero
python do_train.py --config=config_loss_macs_nb101_nas

References

@article{li2021generic,
  title={Generic neural architecture search via regression},
  author={Li, Yuhong and Hao, Cong and Li, Pan and Xiong, Jinjun and Chen, Deming},
  journal={Advances in Neural Information Processing Systems},
  volume={34},
  pages={20476--20490},
  year={2021}
}