Module COMP53715: Deep Learning

Department: Computer Science

COMP53715: Deep Learning

Prerequisites

• None

Corequisites

• None

Excluded Combination of Modules

• None

Aims

• To enable students to design solutions that learn to solve difficult high-dimensional problems across a variety of tasks.
• To position students to understand much of the current literature on deep learning, and be able to extend the knowledge with future study.

Content

• Foundations of machine learning and deep learning: linear regression, Bayesian infer-ence, bias/variance, generalisation, regularisation, cross-validation.
• PyTorch tensor programming and gradient descent.
• Designing architectures: feedforward neural networks, convolutional neural networks, residual architectures, transformers and attention.
• Modelling approaches: sequential models, generative adversarial networks, variational autoencoders, diffusion models, flow models.
• Training and optimisation on GPU servers, model evaluation and inference.

Learning Outcomes

• By the end of this module, students should be able to demonstrate:
• an understanding of the key principles of deep learning for use in curating datasets, designing, training and evaluating models.
• a critical understanding of the foundational approaches within current deep learning literature, neural network architectures and neural network architecture components.
• a systematic understanding of statistical methods and techniques used in machine learning and of statistical learning theory with respect to deep learning approaches.

• By the end of this module, students should be able to demonstrate:
• the scientific approach to design, training, validation, and testing of deep neural net-works using modern deep learning libraries in a broad range of applications.
• an ability to understand and identify inherent issues in dataset and algorithmic bias prior to training or architecture design.
• the ability to design efficient and bespoke neural networks with respect to the task requirements and dataset characteristics.

• By the end of this module, students should be able to demonstrate:
• an ability to critically evaluate and analyse complex problems according to the data structure and characteristics.
• an ability to understand the high-level theory and effectively communicate technical in-formation associated with deep learning literature.
• an ability to utilise modern GPU servers for learning to solve difficult high-dimensional problems.

Modes of Teaching, Learning and Assessment and how these contribute to the learning outcomes of the module

• Lectures enable the students to learn new material relevant to machine and deep learning, as well as their applications.
• Computer classes enable students to acquire necessary coding skills, learn about the relevant libraries and packages and receive feedback on their work.
• The summative assignment assesses the learnt knowledge and application of methods and techniques. It consists of a coding exercise with accompanying report.

Teaching Methods and Learning Hours

Summative Assessment

Formative Assessment:

Via computer classes

■ Attendance at all activities marked with this symbol will be monitored. Students who fail to attend these activities, or to complete the summative or formative assessment specified above, will be subject to the procedures defined in the University's General Regulation V, and may be required to leave the University