sepx-paper

This repository contains all the source codes to reproduce the experimental results reported in paper "Shortest Edit Path Crossover: A Theory-driven Solution to the Permutation Problem in Evolutionary Neural Architecture Search", which is published in ICML 2023. (Arxiv Link: https://arxiv.org/abs/2210.14016)

How to use

Please pip install networkx==2.5 first, then replace networkx/algorithms/similarity.py with the attached similarity.py.

For other dependency requirements and specific guidelines about how to run the source codes for reproducing the experimental results, please see the README files in each individual folder. More specifically: \

• nas-bench-101 includes the guideline and all the source codes for reproducing the main experimental results, which are using NAS-Bench-101 dataset.
• nas-bench-nlp includes the guideline and all the source codes for reproducing the experimental results related to NAS-Bench-NLP dataset.
• nas-bench-301 includes the guideline and all the source codes for reproducing the experimental results related to NAS-Bench-301 dataset.
• expected_improvement_heatmap includes the guideline and all the source codes for reproducing the results of the numerical analysis in Section "Comparisons based on Theory".

Citation

If you use SEPX in your research, please cite it using the following BibTeX entry:

@InProceedings{qiu:icml23,
  title={Shortest Edit Path Crossover: A Theory-driven Solution to the Permutation Problem in Evolutionary Neural Architecture Search},
  author={Qiu, Xin and Miikkulainen, Risto},
  booktitle={Proceedings of the 40th International Conference on Machine Learning},
  pages={28422--28447},
  year={2023},
  volume={202},
  series={Proceedings of Machine Learning Research},
  month={23--29 Jul},
  publisher={PMLR},
}