skip to content

Deconfounding using Spectral Transformations

Presented by: 
Domagoj Ćevid
Peter Bühlmann
Friday 29th June 2018 - 11:45 to 12:30
INI Seminar Room 1
High-dimensional regression methods which rely on the sparsity of the ground truth, such as the Lasso, might break down in the presence of confounding variables. If a latent variable affects both the response and the predictors, the correlation between them changes. This phenomenon can be represented as a linear model where the sparse coefficient vector has been perturbed. We will present our work on this problem. We investigate and propose some spectral transformations for the data which serve as input for the Lasso. We discuss assumptions for achieving the optimal error rate and illustrate the performance on a genomic dataset. The approach is easy to use and leads to convincing results. The talk is based on joint work with Nicolai Meinshausen.
The video for this talk should appear here if JavaScript is enabled.
If it doesn't, something may have gone wrong with our embedded player.
We'll get it fixed as soon as possible.
University of Cambridge Research Councils UK
    Clay Mathematics Institute London Mathematical Society NM Rothschild and Sons