jimburton/COMP0197

UCL Module: Applied Deep Learning

COMP0197: Applied Deep Learning

UCL Module | CS | UCL Moodle Page

Term 2, Academic Year 2022-23

Module Lead
Yipeng Hu yipeng.hu@ucl.ac.uk

1. Development environment

The module tutorials (see bellow) and coursework use Python, NumPy and an option between TensorFlow and PyTorch. The Development environment document contains details of the supported development environment, though it is not mandatory.

2. Tutorials

Quick start

To run the tutorial examples, follow the instruction below.

First, set up the environment:

conda create --name comp0197-tf tensorflow pillow
conda activate comp0197-tf

conda create --name comp0197-pt pytorch torchvision
conda activate comp0197-pt

Additional libraries required for individual tutorials are specified in the readme file in each tutorial directory.

Scripts with "_tf" and "_pt" postfix are using TensorFlow 2 and PyTorch, respectively.

All visual examples will be saved in files, without requiring graphics.

Then, change directory cd to each individual tutorial folder and run individual training scripts, e.g.:

python train_pt.py   

or

python train_tf.py  

Convolutional neural networks

Image classification
Image segmentation

Recurrent neural networks

Text classification
Character generation

Variational autoencoder

MNIST generation

Generative adversarial networks

Face image simulation

3. Reading list

A collection of books and research papers is provided in the Reading List.