Improved Search in Neuroevolution Using a Neural Architecture Classifier With the CNN Architecture Encoding as Feature Vector

This work proposes an improved neuroevolution search strategy that incorporates a neural architecture classifier to reduce the computational cost of evaluating candidate CNN architectures during genetic algorithm-based search. CNN architectures are encoded as feature vectors and used to train a classifier that predicts architecture quality, enabling the search process to skip costly full evaluations for unpromising candidates. The approach demonstrates reduced search time while maintaining competitive performance in the discovered architectures.

How to cite

@article{pilataxi2024neuroevolution,
  author    = {Pilataxi, Jhon I. and Zambrano, Jorge E. and Perez, Claudio A. and Bowyer, Kevin W.},
  journal   = {IEEE Access},
  title     = {Improved Search in Neuroevolution Using a Neural Architecture Classifier With the CNN Architecture Encoding as Feature Vector},
  year      = {2024},
  volume    = {12},
  pages     = {11987--12000},
  doi       = {10.1109/ACCESS.2024.3355804}
}

Recommended citation: J. I. Pilataxi, J. E. Zambrano, C. A. Perez and K. W. Bowyer, "Improved Search in Neuroevolution Using a Neural Architecture Classifier With the CNN Architecture Encoding as Feature Vector," in IEEE Access, vol. 12, pp. 11987-12000, 2024, doi: 10.1109/ACCESS.2024.3355804.
Download Paper | Download Bibtex