Efficient Optimization of Convolutional Neural Networks Using Particle Swarm Optimization

摘要

This work presents methods to automatically find optimal parameter settings for convolutional neural networks (CNNs) by using an evolutionary algorithm called particle swarm optimization (PSO). Even though the parameter space is extremely large (> 10), we experimentally show that a better parameter setting can be found for Alexnet configuration for five different image datasets. We have also developed two candidate pruning algorithms for efficient evolutionary process. In the experiments, we achieved 0.7-5.7% improvements from the original parameter sets in Caffe, while requiring only 2-4% of processing cost of the naive PSO-based approach.