Inference in high dimensional regression

January 20 to January 23, 2015

at the

American Institute of Mathematics, San Jose, California

organized by

Peter Buehlmann, Andrea Montanari, and Jonathan Taylor

Original Announcement

This workshop will be devoted to exploring recent methodological and theoretical advances in inference for high-dimensional statistical models. Classical statistical theory analyzed in great detail the case in which the number of parameters is much smaller than the number of samples. Methods testing statistical hypotheses in this context lie at the foundation of all empirical sciences.

Modern datasets are exceptionally fine-grained and this poses a number of new challenges to the classical framework. Technically, this is the so-called high-dimensional regime whereby the number of parameters is of the same order, or even larger than the number of samples. It is crucial, in this case, to make some structural assumptions regarding the unknown parameter vector -- e.g. imposing sparsity constraints.

While the estimation problem in high-dimension has been object of considerable amount of research over the last 10 years, much less is known about hypothesis testing or confidence intervals. The question is of great practical relevance, and this has spurred several methodological proposals over the last year, in the context of regression modeling.

This workshop aims at bringing together researchers in the field and promote collaboration and discussion around the problem of performing hypothesis testing or computing confidence intervals in high-dimensional statistical problems.

Material from the workshop

A list of participants.

The workshop schedule.

A report on the workshop activities.

Papers arising from the workshop:
Selective sequential model selection
Sure screening for Gaussian graphical models
The p-filter: multi-layer FDR control for grouped hypotheses