Research on the Application of Convolutional Neural Networks on MNIST Datasets

This paper explores the application of Convolutional Neural Networks (CNNs) and Fully Connected Neural Networks (FCNNs) on the MNIST dataset for handwritten digit recognition. The study underscores the significance of digit recognition in various sectors, including automation, accessibility, data validation, security, education, and as a foundation for complex AI systems. The research compares the performance of FCNNs and CNNs, highlighting the latter's superiority due to its ability to capture spatial features and resist overfitting. The paper details the structure and evaluation of both network types, with CNNs achieving a higher accuracy of 99.09% compared to FCNNs at 97.8%.