Probabilistic Machine Learning

Aims

Many machine learning methods can be interpreted as "propose a probabilistic model, and fit it to the dataset". This module will study three such methods in depth: neural networks including classifiers, sequence models, autoencoders, and adversarial training; ranking; and document topic modelling. The common themes will be how to formulate a model, how to fit it, how to evaluate it, and how to reason about uncertainty.

Syllabus

The course will comprise three parts:

1. Neural networks as probabilistic models
2. Probabilistic ranking
3. Topic modelling.

Some of the course will be taught by the Department of Computer Science and Technology and some will be taught via lectures from the Department of Engineering. Please refer to the 'Course Materials' tab for more details.

Assessment

There are four pieces of coursework. The first three are structured exercises designed to reinforce the lectures. The fourth is an open-ended investigation of a topic that you chose from a small list, drawing on the main themes of the lecture course.

• Three exercises, one for each topic (10% each)
• Investigative project (70%)

The investigative project should be written up in the style of a NIPS or ICML conference paper, 8 pages plus one for references.

Further Information

Many of the lectures for this module are delivered by the Department of Engineering at their Trumpington Street site. Students wishing to take this module should note that the Department of Engineering is about 2 miles from the Computer Laboratory.

Due to COVID-19, the method of teaching for this module will be adjusted to cater for physical distancing and students who are working remotely. We will confirm precisely how the module will be taught closer to the start of term.