Evaluating swarm-intelligence algorithms for NAS

I recently concluded and published a work in evolutionary Neural Architecture Search, one of my research interests.

This project analyzed, critically, the potential benefits of using evolutionary or swarm-optimization techniques for optimizing chain-structured neural networks for image classification. I conducted this work because I felt that control experiments were lacking in the Neural Architecture Search field. The results suggest that, at least for the simple search space of chain-structured networks, random search is quite competitive to other, more sophisticated approaches.

You can find the full text here.