Description

This project uses a range of neural network architectures to classify the MNIST handwritten digit and MNIST fashion datasets, acting as a comprehensive comparison of different approaches with informative data visualizations. Included is a hand-coded single perceptron model, multi-layer perceptron architecture, convolutional neural network as well as a multi-task learning CNN model used on the fashion MNIST dataset.

project_notebook.ipynb works through each implementation sequentially using the following python files:

1. pca.py
2. perceptron.py
3. mlp.py
4. cnn.py
5. visualizing_cnn.py
6. multitask_learning.py

All code uses standard python files as well as tensorflow for neural network architectures.

Report.pdf is a comprhensive explanation of the methods used and an analysis of the results.