Hyperparameters’ Optimization of Convolutional Neural Network Using Diversity-Guided Genetic Algorithms for Image Classification

Convolutional Neural Networks (CNNs) are an effective tool for image classification and other computer vision problems. However, getting ideal performance necessitates careful adjusting of hyperparameters, which may be a difficult and time-consuming operation when performed manually. Genetic Algorithms (GAs) are a common approach for hyperparameter optimization; however, they have certain limitations, such as premature convergence and inadequate search space exploration. To overcome these challenges, we present Diversity-Guided Genetic Algorithms (DGGAs), a unique approach for hyperparameter optimisation. DGGAs improve GA performance by introducing diversity preservation mechanisms into the GA architecture. By promoting variation among candidate solutions, DGGAs enable more efficient exploration of the hyperparameter search space, resulting in the discovery of a greater range of high-quality configurations. To test the performance of DGGAs, we performed extensive experiments on the CIFAR-10 dataset, a widely recognized benchmark in computer vision. The experimental results show that our approach achieves competitive accuracy results while maintaining significantly higher diversity among the generated solutions compared to state-of-the-art hyperparameter optimization methods.