PIMS-Institute of Applied Mathematics Distinguished Lecture: Fred Roosta

  • Date: 09/19/2016
  • Time: 15:00
Lecturer(s):
Fred Roosta, UC Berkeley
Location: 

University of British Columbia

Topic: 

UoI Lasso: Union of Intersection Method for Lasso

Description: 

For many modern scientific and machine learning applications, it is often desired to use statistical learning methods that are scalable, highly predictive, highly interpretable, and robust. However, generally speaking, current methods do not simultaneously enjoy all of these properties. We introduce the Union of Intersections (UoI) framework for designing such “all-in-one” methods. In particular, within the context of Lasso, we introduce UoI Lasso and describe its superior theoretical properties under less restrictive conditions. We further demonstrate the performance of UoI Lasso on a variety of benchmark biomedical data: extraction of meaningful functional networks from human electrophysiology recordings and dramatically more parsimonious prediction of behavioral and physiological phenotypes from genetic data.

Other Information: 

Location: ESB 2012

Note for Attendees: Tea served before the talk in ESB 4133 (PIMS lounge)