Peer Review Request - Acceptance Confirmation
BMS-EENG-2024-207
Article Title:
Comparative Analysis of CNN Performances Using CIFAR-100 and MNIST Databases: GPU vs. CPU Efficiency
Abstract:
Currently, the use of Convolutional Neural Networks (CNN) and in particular deep learning algorithms has revolutionized image processing and classification techniques. This article presents a comparative analysis of the performances of three neural network architectures: VGG-16, ResNet-50 and ResNet-18, focusing specifically on training time. The study also evaluates test accuracy and training loss on CIFAR-100 and MNIST datasets. The experiments, carried out under Ubuntu 22.04 with the PyTorch Framework, highlight well-established significant differences in performance using an NVIDIA-GPU or an Intel-CPU. The results highlight the substantial advantages of GPUs for deep learning algorithms, particularly for image classification, and emphasize the importance of properly selecting the right architecture and adequate Material in order to optimize performance and efficiency in artificial intelligence applications.