FMeCo-SWOT

J. Huang, B. Xue, Y. Sun, M. Zhang, and G. G. Yen, “Activation features empowered: An efficient gradient-free proxy to zero-shot neural architecture search,” Pattern Recognition, vol. 179, p. 113917, 2026.

📑 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 ./datasets
• Download benchmark datasets and put them in ./APIs
  • NAS-Bench-201: https://drive.google.com/file/d/16Y0UwGisiouVRxW-W5hEtbxmcHw_0hF_/view
  • NDS: https://dl.fbaipublicfiles.com/nds/data.zip
• Evaluate DPP on NAS-Bench-201 and NDS by running sh run.sh

Citation

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

@ARTICLE{FMeCo-SWOT,
  author={Huang, Junhao and Xue, Bing and Sun, Yanan and Zhang, Mengjie and Yen, Gary G.},
  title={Activation features empowered: An efficient gradient-free proxy to zero-shot neural architecture search},
  journal={Pattern Recognition},
  year={2026},
  volume={179},
  number={},
  pages={113917},
  issn={0031-3203},
  doi={https://doi.org/10.1016/j.patcog.2026.113917}}