“MPSO: Mutation-Enhanced Particle Swarm Optimization for CNN Architecture Optimization”

This study introduces a novel approach, MutationEnhanced Particle Swarm Optimization (MPSO), for autonomously optimizing convolutional neural network (CNN) architectures. The MPSO method integrates the Mutation operator with particle swarm optimization, enabling targeted variations in CNN architecture elements such as layer counts and filter sizes. Through extensive experimentation on benchmark datasets, MPSO demonstrates superior performance compared to alternative methods, showcasing its effectiveness in streamlining CNN design without human intervention. The strategic mutation strategy employed in MPSO facilitates exploration within the search space, leading to the discovery of more effective network designs, particularly in image classification tasks. Future research aims to validate MPSO on larger datasets like ImageNet and explore its adaptability to newer CNN architectures, ultimately advancing autonomous CNN scaling and optimization methodologies.