CS826 - Deep Learning Theory and Practice

TIMETABLE	TEACHING MATERIAL
Credits	20
Level	5
Semester	Semester 1
Availability	Mandatory
Prerequisites	N/A
Learning Activities Breakdown	Lectures: 20 hours \| Labs: 20 hours Homework / Private Study: 160
Assessment	One group assignment is worth 40%, and classes/lab tests or quizzes are worth 60%.
Lecturer	Andreas Neofytou

The aim of this module is to endow students with:

an understanding of the key algorithms and techniques with deep learning.
an understanding of the limitations of the current technologies and their future trend.

After completing this module participants will be able to:

Understand the challenges in the training of deep neural network and how to overcome the challenges with few shot learning techniques.
Understand how to process sequential data with deep neural network and the applications in natural language processing.
Develop understanding of the potentials and generalisation of deep neural networks through theoretical analysis.
Understand deep generative models and their main approaches.
Understand how to build Deep Neural Networks in Python using packages such as PyTorch and implement such networks.

Review of fundamentals and basics (neural networks, math).
Training models in deep learning with small examples and approaches for lifelong learning (i.e. transfer Learning, metal learning, few shot learning).
Attentions and transformers in deep learning, their relations with recurrent neural network, LSTM and GRU, and their applications (e.g. language translation and sentimental analysis).
Universal Approximation Theorem, the potentials of deep neural networks, the major challenges in the training and their generalisations.
Unsupervised learning and generative models (e.g. flow based models, variational autoencoder and generative adversarial networks, their pros and cons).

Syllabuses - PG