SAPSONAS: Surrogate assisted PSO for Neural Architecture Search

Recently, to accelerate the search for neural architecture, surrogate-assisted evolutionary algorithms (SAEAs) have been shown to perform well in optimizing computationally complex problems (EOPs ). In this article, the infill criterion is used to make SAEAs effective and also to have a more efficient neural architecture search use network embedding. Architectures with similar structures are closer to each other in the embedding space. Due to the use of the evolutionary algorithm of particle swarm optimization(PSO), the maximum accuracy in the database can be reached in fewer generations. Experiments on three search spaces with different dimensions from NASBench show the superiority of the proposed SAPSONAS method.

کلیدواژه ها:

Keywords—SAEA ، EOP ، embedding space ، NAS ، NASBench ، surrogate model