Course Content
In this course, we will take a look at Machine Learning in order to consolidate the theoretical, algorithmic and practical basics.
Learning methods based on bagging, boosting and stacking will then be presented. Following this, and depending on the time remaining, short introductions to the following topics will be given: PAC-Bayes theory, which is a very suitable theoretical framework for ensemble methods, metric learning and domain adaptation.
The aim of this course is to assimilate good experimental practices in order to be able to compare methods correctly. It also aims to present different aspects of Machine Learning via set methods.
Lectures
Below, you will find the different slides used for the different lectures and a complete note which contain further information on all the topics studied during the different sessions.
Exercises
The following file contains a small list of exercises dealing with ensemble methods and PAC-Bayesian learning. In some of them, we only expect you to do some computations in order to check if you have well understood how the different methods work. The remaining exercises are more theoretical.
Practical Sessions
You will find the different practical sessions associated with each part of the course. Unfortunately we will not be able to deal with all of them during the sessions but you can do them on your side, at home
The first part contains main files and datasets used for all the practical sessions
TP 1 - An experimental protocol for supervised learning
TP 2 - Bagging and Random Forest
TP 3 - Generalities around Boosting: Adaboost and Gradient Boosting