At the end of this course, you will be able to demonstrate knowledge of the following theoretical concepts:

Foundations of Bayesian networks and other types of probabilistic graphical models.

Model testing, statistical equivalence, and parsimoniousness (Occam's razor).

Message passing and other foundations of probabilistic inference.

Key causal inference concepts such as intervention and confounding.

Further, you will have acquired the following practical skills:

Build a Bayesian network model of a problem domain that you are familiar with.

Use inference algorithms for probabilistic reasoning in Bayesian networks.

Statistically evaluate the model fit of a Bayesian network to a given dataset.

Use structure learning algorithms to generate plausible network structures for given datasets.

Use a Bayesian network to help answer causal inference questions from observational data -- that is, learn how to tackle questions such as "will getting a Master's degree increase my future salary?"

Content

Bayesian networks are powerful, yet intuitive tools for knowledge representation, reasoning under uncertainty, inference, prediction, and classification. The history of Bayesian Networks dates back to the groundbreaking work of Judea Pearl and others in the late 1980s, for which Pearl was given the Turing Award in 2012.

Bayesian networks are used in many application domains, notably medicine and molecular biology. This course will cover the necessary theory to understand, build, and work with Bayesian networks. It will also introduce how Bayesian networks provide a much needed foundation for causal inference, giving rise to what is sometimes called the "causal revolution".

Instructional Modes

Lecture

Tutorial

Self-study

Level

Presumed foreknowledge

Previous knowledge in probability and statistics may help to digest the content more quickly, but the necessary bits such as the basic laws of probability and conditional probability will also be recapped at the beginning of the course. Since course includes two assignments, previous knowledge of python, R, or both can be beneficial, but several students without prior knowledge in R were able to pick up enough R skills during the course to complete the assignments successfully.

Test information

Written exam, two project reports & code, including a presentation.
The exam contributes 50% to the final grade; the projects 25% each.

In accordance with the Rules and Regulations, the grade for the written exam should be a minimum of 5.0.

Specifics

The course is part of the Computing Science Data Science theme and also part of the AI master programme.