¹û¶³Ó°Ôº

Aims:

The aims of the module are to provide familiarity with Bayesian approaches to deep learning, currently a very active area of research.

Intended learning outcomes:

On successful completion of the module, a student will be able to:

1. Identify both the benefits and difficulties posed by Bayesian approaches deep learning.
2. Implement a number of different approaches to Bayesian neural networks for classification and regression and understand the trade-offs between methods.
3. Recognize the key concepts related to uncertainty quantification in predictive models.
4. Implement approaches to using deep learning for generative modelling and unsupervised learning.
5. Identify the fundamental connections between both deep learning and probabilistic modelling.

Indicative content:

This module covers topics on Bayesian approaches to deep learning, including theory and applications. We focus on understanding the role of uncertainty in deep learning models in both supervised and unsupervised settings.

The following are indicative of the topics the module will typically cover:

• Bayesian treatments of model parameters.
• Uncertainty quantification.
• Inference in deep generative models.
• Applications which benefit from a Bayesian approach, including active learning and semi-supervised learning.

Requisites:

To be eligible to select this module as an optional or elective, a student must: (1) be registered on a programme and year of study for which it is formally available; (2) have a strong understanding of Linear Algebra, Multivariable Calculus, Probability Theory; and (3) have taken at least one introductory machine learning module, for example Supervised Learning (COMP0078) or Introduction to Machine Learning (COMP0088) (or be concurrently enrolled in such a module).