Eliminating Network Depth: Genetic Algorithm for Parameter Optimization in CNNs

Recent advances in Convolutional Neural Networks (CNNs) have significantly enhanced image classification performance. However, CNNs often require large numbers of parameters, leading to increased computational complexity, prolonged training times, and substantial resource demands. Achieving higher classification accuracy typically involves deepening network architectures, which further exacerbates these challenges. This paper proposes a novel method based on a genetic algorithm to optimize parameter selection, enabling the construction of CNNs that achieve superior accuracy with fewer parameters. By focusing on parameters with the most significant impact on performance, the method reduces the need for deeper networks, thereby minimizing computational costs. Experimental results demonstrate that the proposed algorithm outperforms its counterparts. For instance, the generated CNN achieves an accuracy improvement of 0.75 percentage points over ResNet-110 while using 84% fewer parameters. These findings highlight the method’s potential to balance efficiency and accuracy, making it a promising solution for resource-constrained applications.