Eugen Pircalabelu, ISBA - UCLouvain

Introduction to high-dimensional statistics

Content:
More and more applications and problems nowadays require the estimation of many parameters. The field of high-dimensional statistics uses data whose effective dimension (read the number of parameters to be estimated) is much larger than dimensions considered in classical multivariate low-dimensional analysis. More than that, the number of parameters to be estimated may be larger than the available sample size, making classical low-dimensional techniques unusable in such settings as they cannot produce consistent estimators when the sample size is small compared to the number of parameters to be estimated.
We introduce in this course the theory of regularized methods in high-dimensional statistics. We address first estimation and prediction problems, move to feature selection problems in high dimensions and concentrate then on tuning parameter procedures. We then present the challenges of statistical inference in this framework. After having acquired this knowledge, we tackle the high-dimensional graphical modeling framework.
Themes investigated:
1.Challenges concerning high-dimensional models
2.Regularized methods in high-dimensional statistics
3.Parameter estimation
4.Tuning parameter selection
5.Feature selection
6.Graphical modeling
7.High-dimensional inference

Volume : 15h, ECTS: 3

Registration:
For UCLouvain members : via Moodle UCLouvain
For non UCLouvain members : please confirm your registration to Prof. E. Pircalabelu directly and do not forget to mention your name, surname, affiliation, and if you will need a certification of participation : Eugen Pircalabelu

Venue and schedule
ISBA C045 : Voie du Roman Pays 20 - 1348 Louvain-la-Neuve

• Monday, February 4, 2019
  08:30-10:30 and 10:45-12:45

• Monday, February 11, 2019
  08:30-10:30 and 10:45-12:45
  and 16:15-18:15

• Monday, February 18, 2019
  08:30-10:30 and 10:45-12:45