Train a Convolutional Neural Network on the Fashion MNIST dataset using PyTorch
Let's train a CNN model on the famous Fashion MNIST dataset. We start with a simple CNN with 2 convolutional layers and 2 fully connected layers which results in a 85% accuracy score. We then optimize it using 3 convolutional layers and 2 fully connected layers, applying padding and dropout at 50%, applying batch norm for better performance, and also reducing the size of the filters and increase the number of filters to have more opportunities for extracting simple features from the input images. The new model has a 90% accuracy score.