EvoREP: Evolving Reliable Ensemble of Proxies for Zero-Shot Neural Architecture Search

J. Huang, B. Xue, Y. Sun, M. Zhang, and G. G. Yen, “EvoREP: Evolving Reliable Ensemble of Proxies for Zero-Shot Neural Architecture Search,” IEEE Transactions on Evolutionary Computation (Early Access), pp. 1–15, 2025. DOI: 10.1109/TEVC.2025.3646736

📑 Read the Paper

Preparation

This code is tested with Python 3.12.7, PyTorch 2.5.1, and CUDA 12.7.

• Download datasets (CIFAR-10, CIFAR-100, ImageNet-16-120) from https://drive.google.com/drive/folders/1T3UIyZXUhMmIuJLOBMIYKAsJknAtrrO4 and put them in ./data
• Download benchmark datasets and put them in ./APIs
  • NATS-Bench-TSS (NAS-Bench-201): https://drive.google.com/file/d/16Y0UwGisiouVRxW-W5hEtbxmcHw_0hF_/view
  • NATS-Bench-SSS: https://drive.google.com/file/d/1scOMTUwcQhAMa_IMedp9lTzwmgqHLGgA/view
  • NDS: https://dl.fbaipublicfiles.com/nds/data.zip

Proxy Search

• Conduct proxy search on NATS-Bench and NDS by running sh run.sh

Citation

If you use this code in your research, please cite the following paper:

@ARTICLE{EvoREP,
  author={Huang, Junhao and Xue, Bing and Sun, Yanan and Zhang, Mengjie and Yen, Gary G.},
  journal={IEEE Transactions on Evolutionary Computation}, 
  title={EvoREP: Evolving Reliable Ensemble of Proxies for Zero-Shot Neural Architecture Search}, 
  year={2025},
  volume={},
  number={},
  pages={1-15},
  doi={10.1109/TEVC.2025.3646736}}