Accelerating Gene-pool Optimal Mixing Evolutionary Algorithm for Neural Architecture Search with Synaptic Flow

Abstract

This study experiments the integration of the zero-cost proxy metric Synaptic Flow with the Gene-pool Optimal Mixing (GOM) crossover to efficiently generate new candidates during an evolutionary neural architecture search (ENAS). Our experiments demonstrate that the Gene-pool Optimal Mixing Evolutionary Algorithm (GOMEA) with Synaptic Flow can obtain top-performing architectures with a small additional overhead compared to a classic Genetic Algorithm. Code is available at: https://github.com/ELO-Lab/SF-GOMENAS.