Hypothesis testing for high-dimensional LASSO models.

Details

Supervisors

Pircalabelu, Eugen

Degree label

Master [120] en statistique, orientation générale, à finalité approfondie

Abstract

Nowadays, available information is increasing and high-dimensional models are more and more common. In this context, obtaining tools for inference in the cases where the number of variables exceeds the number of observations is a crucial task. In this work, we compare methods for debiasing the LASSO estimator that allow the construction of high-dimensional hypothesis testing. We inspect these estimators from a theoretical point of view but also through simulations and applications on real data. The extension to GLMs is also discussed. We propose a model averaging estimator based on these desparsified estimators and we inspect its performances through a small simulation study.