The programme runs from January to December. The year is split into three blocks:
- Block 1: core modules
- Block 2: elective modules
- Block 3: research project
Students must complete the four 15-credit core modules, six of the 10-credit elective modules on offer, and the 60-credit research project, for a programme total of 180 credits.
The programme requires one year of intensive full-time study, or two years of part-time study. Prospective part-time students should consult the programme coordinator on how best to split their modules and project over the two years.
Block 1: core modules (15 credits each)
Mathematics for Machine Learning
A review of fundamental mathematical concepts as tools of the ML and AI trade, covering relevant topics from linear algebra, multivariate calculus, optimisation, and mathematical statistics.
Probabilistic Modelling and Reasoning
Probability theory, marginalisation, sum-product decomposition, Markov blankets, classic hidden Markov models, expectation maximisation, probabilistic graphical models, data completion, and information theory.
Foundations of Deep Learning
The basics of deep learning as a precursor for more advanced modules. A recap of ML fundamentals, followed by topics specific to deep neural networks e.g. gradient-based training, common architectures, autoencoders and generative models.
Applied Machine Learning at Scale
This module looks at how ML is applied to internet-scale systems. Topics include A/B testing, ranking, recommender systems, modelling of users and online content (like news stories), and network effects.
Block 2: electives (10 credits each; must complete six)
Offering of these modules will depend on lecturer availability, and not all of them are necessarily available every year. The list for 2024 will be confirmed soon.
A study of convolutional neural networks for popular computer vision applications like image classification, object detection, semantic segmentation, style transfer, image captioning, and image generation.
Natural Language Processing
Word embeddings, part-of-speech tagging and syntactic parsing, topic modelling, language modelling with recurrent and convolutional neural networks, machine translation with seq2seq models and attention, and sentence classification.
Reinforcement Learning and Planning
Decision and control theory, exploration, Q-learning and policy gradients, hierarchical RL, Markov decision processes, model-based RL, multi-agent RL, planning and navigation.
Techniques to model and predict temporally varying data (e.g. financial data, weather, audio or video). We look at state-space models, hidden Markov models, recurrent neural networks, long short-term memory gates, and the Transformer model.
Advanced Probabilistic Modelling
Restricted Boltzmann machines; approximate inference techniques for PGMs like belief propagation and variational inference; Bayesian non-parametric models such as Gaussian and Dirichlet processes, and collapsed Gibbs sampling.
Optimisation for Machine Learning
Convex optimisation in high-dimensional space, stochastic mini-batch gradient descent, momentum, Nesterov theory, and practical methods like RMSProp and Adam. More advanced topics like conjugate exponentials and stochastic variational inference may also be covered.
Monte Carlo Methods
Approximation techniques for statistical inference in ML. Metropolis-Hastings, Gibbs sampling, importance sampling, slice sampling and exact sampling. Techniques for normalising constants, annealing and thermodynamic integration may also be covered.
Artificial Intelligence and the Brain
A look at topics from neuroscience that inspire AI research: Hebbian learning, simple perceptrons, the canonical microcircuit, predictive coding, dopamine-coded reward prediction, hippocampal feedback, the visual hierarchy and the auditory processing system.
Advanced Topics I and II
To keep up with progress in ML and AI research, we reserve two elective modules for advanced topics that do not fit naturally into any of the other modules of the programme.
Block 3: research project (60 credits)
Students work on individual projects with guidance from supervisors. The project entails forumulating objectives, surveying the literature, applying what was learned in the modules, and evaluating results critically. The output will be a high-quality conference or journal style paper and oral presentation.