Master2 Data Sciences

Theoretical guidelines for high-dimensional data analysis

These lectures are offered in the M2 Data Sciences

Schedule

28/11: room 0A1, Institut de Mathématiqued d'Orsay, 14h-18h
05/12: room 0A1, Institut de Mathématiqued d'Orsay, 14h-18h
12/12: amphi Yoccoz, Institut de Mathématiqued d'Orsay, 14h-18h
19/12: room 0A1, Institut de Mathématiqued d'Orsay, 14h-18h
09/01: room 0A1, Institut de Mathématiqued d'Orsay, 14h-18h

Access

The Institut de Mathématiques d'Orsay is located in Building 307, Orsay campus, 5min walk from RER station Orsay-Ville.
Access plan

Warning!

This course is intended as a theoretical reflexion on the practice of data analysis. It is not suited as a training for a PhD in mathematical statistics. You should instead follow the course Statistiques en grande dimension offered in the Master2 StatML and PS and MSV.

Program

Goal of the lectures:

to draw your attention to some issues in data analysis, and some proposals to handle them;
to learn to read a research paper, to catch the take home message and to identify the limits;
to favor your own critical analysis of research papers.

The lecture will be based on some recent research papers (How to read a paper?). The list of the research papers below is subject to change before the start of the lectures. The presence during the lectures is mandatory and taken into account in the final evaluation.

Lecture	Topic	Paper(s)	Slides	Further reading
1	Strength and weakness of the Lasso	Paper	Slides	No free computationnal lunch
2	False discoveries, multiple testing, online issue and link to bandits problems	Paper	Slides	Reliability of scientific findings? Quality Preserving Databases? Online FDR control
3	Adaptive data analysis	Paper	Slides	Kaggle overfiting
4	Curse of dimensionality. When interpolation meets underfitting	Paper	Slides and Lecture notes	Robust PCA
5	Robust learning	Paper	Slides	Learning with Median Of Means

Evaluation

The reports must be sent by email by February 15 in a zip file including:
- the report in pdf format (10 to 20 pages)
- the source code for the numerics
Look at the instructions for your report!