Comparison of optimization algorithms based on swarm intelligence applied to convolutional neural networks for face recognition

Abstract

In this work, a comparison of optimization techniques based on swarm intelligence to design Convolutional Neural Networks is performed. The optimization techniques used in this comparison are Grey Wolf Optimizer (GWO) and Whale Optimization Algorithm (WOA). The algorithms design convolutional neural networks (CNNs) architectures applied to face recognition. These techniques were chosen due to their similarity in their processes to find optimal results, such as their searching of prey. The design of CNNs consists of the number of layers (convolutional and fully connected), number and size of the filters, neurons fully connected, batch size, epoch, and algorithm for the learning phase. The simulation results are compared, using a different number of images for the learning phase to know which technique has a better performance using a smaller number of images to CNN design.