This course is a follow up to Martin Weigt’s « Computational science » class.
The class will consist in lectures and lab sessions where we will code some algorithms from scratch, and apply them to textbook or real-life datasets.
We will cover the basic building blocks of modern ML, emphasizing the mathematical origins of the algorithmic recipes used in learning algorithms, which allows to make sense out of the diversity of algorithms in use. In the same time, key methodological points will be outlined and illustrated on concrete examples (datasets and tasks). Typical success and pitfalls of statistical learning and their interpretation by statistical physics methods will be provided.
Overall, we will cover the vocabulary used in ML litterature, so as to become autonomous when confronted with new methods.
Depending on the appetite of the class, we will spend more or less time on Deep Learning methods.