The “relax and split” approach has proven to be practical and versatile for variable selection, allowing easy implementation and better support detection while using any sparsity promoting regularizer. In this talk, we review the basic idea – combining relaxation and partial minimization – and show results for linear models as well as more recent work in the context of mixed effects, a key nonlinear regression setting used to analyze repeated measurements. We also discuss ongoing work in the RS theme. 

Refreshments will be served preceding the talk, starting at 2:45.

We gratefully acknowledge funding support from the Pacific Institute of Mathematical Sciences (PIMS).