Exploration of Genetic Algorithm-Driven Hyperparameter Optimization for Convolutional Neural Networks

Abstract

Convolutional Neural Networks (CNNs) have been widely used in artificial intelligence (AI) research. However, the time-intensive nature of hyperparameter optimization remains a major challenge. In this study, we investigated three distinct CNN architectures, namely, ResNet, Multi-class, and ConvLSTM, by employing Genetic Algorithm (GA). The GA-driven optimization process efficiently adapts the parameters, resulting in significant improvements in validation accuracy and reduced training time. Moreover, our findings indicate that compared to exhaustive manual exploration, GA substantially reduces the time required for hyperparameter optimization, thereby offering a more efficient and effective approach for enhancing model outcomes in reasonable timeframes. This underscores the importance of employing advanced optimization techniques such as GA in AI research to achieve superior model outcomes.