UvA Deep Learning Course 2020 - Practicals

Overview

This repo contains my solutions to the 3 assignments of the course Deep Learning realised at the Univeristy of Amsterdam in Fall 2020. Among others, here you can find PyTorch implemenations and/or theory of the following:

Assignment 1 (score 96/100):

• MLP (additionally coded in NumPy from scratch)
• Custom module for Layer Normalization
• CNN VGG13 architecture with skip connections.

Assignment 2 (score 105/100):

• Vanilla LSTM
• bi-LSTM
• Application: LSTM for text generation
• Graph Neural Networks (theory only)

Assignment 3 (score 95/100):

• VAE
• GAN
• Flow-based models (theory only)

Next to the code, under each assignment folder one can find my report (Report.pdf) outlining the theory and obtained results. If you're interested, here is the official github repository of the course including past versions of assignments.

Environments

A conda environment called dl2020 is provided. It contains all packages needed for the course (practicals and tutorials). For your own computer, use the file environment.yml to create the environment. For your account on Lisa (UvA supercomputer), please use the file environment_Lisa.yml which installs the environment dl2020 with CUDA 10.1 support.