Statistical learning with extremes,
Master MVA, université Paris-Saclay
Joint course with Stephan Clemencon, Antoine Doize.
Location/Schedule in 2025: at Telecom Paris (Palaiseau), on Thursdays, 9:30--12:30, rooms: following this schedule
Registration (mandatory) for 2025
Course material
Homework
Homework should be sent by email to statlearn.extremes.mva@gmail.com (only).
-
Questions related to Lab 1 (due by 1/11/2025), see instructions received by email).
UPDATE: corrected typos in the assignment: updated questions sheet -
(due by 21/11/2025) practical + theoretical questions. Exercises 1 and 2 below could be handed out as scanned, readable, handwritten notes.
- Questions related to lab2
- Exercise 1 : prove Proposition 3.2.2 from the lecture notes
- Exercise 2: (Exercise 3.4 from the lecture notes) prove the integal representation for the limit measure of multivariate extremes, invloving the angular measure
Guidelines for oral presentations and report, list of papers
- Instructions (last updated 2025/10/27)
Recommended reading
The second part of this course focuses on statistical learning theory for extreme value analysis. A convenient reference is the survey paper co-written by Stephan Clemencon and myself (working paper).It is highly recommended to familiarize yourself first with basic statistical learning theory. To do this, take Nicolas Vayatis' course 'Introduction to Statistical Learning'.
As for general references, good starting points are
- The book Foundations of Machine Learning (Mohri et al.), Chapters 2,3;
- G. Lugosi's
lecture notes on the topic of binary classification. The
first four sections are particularly relevant to our context.
N.B. These lecture notes are part of the book `Principles of Nonparametric Learning' .