Enhanced Convolutional Neural Networks for MNIST Digit Recognition

Abstract

This study addresses the ongoing pursuit of achieving optimal performance in digit recognition tasks, focusing on the widely studied MNIST dataset. Our motivation stems from the challenge of accurately classifying the remaining 1% of images, despite the relatively high 99% accuracy achieved by existing models. In this work, we present a simplified approach to convolutional neural network (CNN) architecture, aiming to streamline model complexity while maintaining or even enhancing performance. Unlike previous approaches, our methodology involves utilizing only two CNN layers with fewer filters, resulting in a reduction in model parameters and learning time. Through rigorous experimentation and evaluation, we demonstrate that our streamlined CNN architecture yields competitive results. Our findings underscore the importance of exploring alternative model architectures and optimization techniques to achieve state-of-the-art performance in digit recognition tasks.