Build a Fashion-MNIST CNN with PyTorch

This document details the process of developing and optimizing a Convolutional Neural Network (CNN) for the Fashion-MNIST dataset using PyTorch. Building upon the foundational work of "Let's Build a Fashion-MNIST CNN, PyTorch Style" [1], this project introduces significant enhancements, including a redesigned CNN architecture, a systematic hyperparameter tuning process, and comprehensive testing and visualization. The optimized model achieves a training accuracy of 99.02% and a test accuracy of 91.01%, a substantial improvement over baseline models. This report outlines the methodology, model architecture, optimization techniques, and data analysis that led to these results.

Read More