Deep learning is an important technology underlying much modern progress in machine learning and artificial intelligence more generally. This module will provide students with a practical grounding in modern neural networks. Starting with some underlying mathematical preliminaries, the module will progress on to discussing the basic structure and operation of differentiable computational networks. Students will then be introduced to a selection of network architectures, and the strengths and weaknesses of each will be explored.
The module is practical in nature, emphasising implementation, training and inference using real-world datasets and the effective and idiomatic use of modern deep learning libraries.
- Foundations of deep learning
- Logistic and linear regression
- Computational graphs and differentiable computing
- Backpropagation
- Basic deep learning
- Feedforward networks
- Training algorithms (gradient descent, ADAM, etc.)
- Activation functions
- Convolutional networks
- Recurrent networks
- Advanced deep learning
- LSTMs
- Transformers
- Generative AI
- Diffusion Networks
- Other architectures dependent on time limits
Basics
1. Logistic and linear regression 2. Computational graphs and differentiable computing 3. Backpropagation
Basic deep learning
1. Feedforward networks 2. Training algorithms (gradient descent, ADAM, etc.) 3. Activation functions 4. Convolutional networks 5. Recurrent networks
Advanced deep learning
1. LSTMs 2. Transformers 3. Generative AI 4. Diffusion Networks5. Other architectures dependent on time limits
This module will use a range of learning and teaching methods including laboratory work, lectures, tutorial sessions, and regular assessments to build up students' confidence in the development of Deep Learning systems.
| Module Content & Assessment | |
|---|---|
| Assessment Breakdown | % |
| Formal Examination | 50 |
| Other Assessment(s) | 50 |