An Introduction to Probabilistic Machine Learning

Probabilistic machine learning has gained a lot of practical relevance over the past 15 years as it is highly data-efficient, allows practitioners to easily incorporate domain expertise and, due to the recent advances in efficient approximate inference, is highly scalable. Moreover, it has close relations to causal inference which is one of the key methods for measuring cause-effect relationship of machine learning models and explainable artificial intelligence. This openHPI course will introduce all recent developments in probabilistic modeling and inference. It will cover both the theoretical as well as practical and computational aspects of probabilistic machine learning.

This course requires Julia programming; we will use the CodeOcean feature of openHPI. We will also assume that the participants have a solid understanding of analysis and calculus.

Week 1:

This will will cover the following four topics: (1) What is Machine Learning, (2) The Role of Probability in Machine Learning, (3) Introduction to Probability Theory and (4) Probability Distributions.

Week 2:

This week will cover the following three topics: (1) Graphical Models: Bayesian Networks, (2) Graphical Models: Factor Graphs and the Sum-Product Algorithm, and (3) Bayesian Ranking (TrueSkill).

Week 3:

This week will cover the following two larger topics: (1) Bayesian Linear Regression, and (2) Gaussian Processes.

Week 4:

This final week will cover these three topics: (1) Bayesian Classification algorithms, (2) Non-Bayesian Classification learning algorithms, and (3) modelling text and image data.