Mathematical Foundations of Deep Neural Networks (MFDNN)

This repository contains the original problems alongside my submissions for Mathematical Foundations of Deep Neural Networks [M1407.001200] taken at Seoul National University (SNU) in the Spring semester of 2024.

All material (assignments themselves, solutions, starter code) is credited to Ernest K. Ryu (https://ernestryu.com/) and available freely on his website.

Included files

. contains the problem sheets themselves alongside the starter code provided for each problem.

./hw_workings contains a Jupyter notebook with my own submissions for the programming questions of each assignment (hw*_code.ipynb) alongside a pdf version of their output (hw*_code_output.pdf). Also included is a pdf of the handwritten workings for the mathematical problems (hw*_workings.pdf).

./hw_solutions contains the solution files organised by assignment number.

Note that dataset files are not included (including mnist, cifar10, and kmnist). The given path variables in the notebooks should be edited to a local copy of these datasets or simply redownloaded using the provided pyTorch datasets loaders (with download=True)

Dependencies

All of my own notebooks were originally created and run on MacOS (arm) with the following versions. mps is used as the device for pytorch where applicable.

Python version: 3.11.8 | packaged by conda-forge | (main, Feb 16 2024, 20:49:36) [Clang 16.0.6 ]
Numpy version: 1.26.4
Matplotlib version: 3.8.0
PyTorch version: 2.2.1