Grey Wolf Optimizer to the Hyperparameters Optimization of Convolutional Neural Network with Several Activation Functions
Yiğit Çağatay Kuyu, Nurettin Ozekmekci · 2022 International Symposium on Multidisciplinary Studies and Innovative Technologies (ISMSIT) · 2022
Hyperparameter tuning has gained a remarkable research interest since identifying a right configuration of hyperparameters gives an accurate model in deep learning. Convolutional neural network (CNN) is one of the most popular deep learning techniques that is widely used for image classification tasks. Obtaining the suitable hyperparameter configuration of CNN is challenging due to the mixed-variable types of hyperparameters and their high number of combinations in the search space. Manual search based on a trial-and-error approach is a time-consuming process and requires professional knowledge. Therefore, an automated search method is needed to reduce human efforts and to design a model with minimal human expertise. Motivated by these, in this study, the grey wolf optimizer algorithm is proposed to automize the process of finding hyperparameters of CNN, and several activation functions are adapted to the proposed framework to highlight the best CNN architecture. The experiments are conducted on the famous MNIST dataset, and comparative analyses are carried out among the models having different activation functions on the same problem.