Machine learning

Course goals

Upon completing this course, the student can

1. relate the behaviour of machine learning methods to their mathematical formulations
2. compute the central quantities of Bayesian machine learning
3. discuss advantages and disadvantages of different computational algorithms used in machine learning
4. implement some machine learning methods in terms of elementary operations in a programming language
5. evaluate the results of a machine learning method.
6. identify alternative explainable AI methods, and choose an appropriate method based on the model family.

Assessment
Assignments (30% of the grade); two exams (35% each), average exam grade must be at least 5.5 to pass.

You qualify for a repair exam if your overall average at that point is insufficient but you have participated in all graded components of the course, your average exam grade is at least 4, and you have made a sufficient effort to do the theoretical homework exercises.

There is no retake opportunity for the assignments.

Prerequisites
For CKI students, this course builds on the first-year course KI2V20001 Introduction to Machine Learning, so the material treated there will be assumed known.

CS students, you should be familiar with the material of the following courses:

• INFOGR Graphics (for linear algebra);
• INFOB3DAR Data analysis and retrieval (for basic concepts of machine learning);
• INFOB3CI Computational intelligence (for probability theory).

In addition, as a CS student you may need to do some additional reading and studying.

Finally, for all students, programming experience with Python and Numpy is highly recommended.

 

Content

In this advanced course about machine learning, we cover several methods for supervised and unsupervised learning, including support vector machines, kernel methods, and advanced techniques used in deep learning.
We will go into the underlying mathematical theory, and the precise way in which certain algorithms operate, such as the backpropagation algorithm used to train neural networks.

Topics:
- Bayesian machine learning: generative models; inference algorithms
- Support vector machines
- Kernel methods
- Unsupervised learning: clustering; principal component analysis
- Latent variable models and variational Bayes
- Explainable machine learning
- Deep learning

Course form
Lectures and tutorial sessions (attendance mandatory). Some of the lectures will be taught in Dutch, the remainder will be taught in English.

Literature

• Simon Rogers & Mark Girolami, "A First Course in Machine Learning", second edition. Chapman and Hall/CRC. Hardback (2016) ISBN 9781498738484, paperback (2020) ISBN 9780367574642, e-book (2016) ISBN 9781315382159.
• additional online material