Adaptive Dimensional Gaussian Mutation of PSO-Optimized Convolutional Neural Network Hyperparameters

The configuration of the hyperparameters in convolutional neural networks (CNN) is crucial for determining their performance. However, traditional methods hyperparameter configuration, such as grid searches and random searches, are time consuming labor intensive. optimization CNN a complex problem involving multiple local optima that poses challenge particle swarm (PSO) algorithms, which prone to getting stuck achieving suboptimal results. To address above issues, we proposed an adaptive dimensional Gaussian mutation PSO (ADGMPSO) efficiently select optimal configurations. ADGMPSO algorithm utilized cat chaos initialization strategy generate initial population with more uniform distribution. It combined sine-based inertia weights asynchronous change learning factor balance global exploration exploitation capabilities. Finally, elite was improve diversity convergence accuracy at different stages evolution. performance compared five other evolutionary including PSO, BOA, WOA, SSA, GWO, on ten benchmark test functions, results demonstrated superiority terms value, mean standard deviation. then applied LeNet-5 ResNet-18 network models. MNIST CIFAR10 datasets showed achieved higher generalization ability than PSO-CNN, LDWPSO-CNN, GA-CNN.