An Interior Point Approach for Data Science, Sasha Aravkin

Tuesday, October 20, 2015, 4:00 pm
GUG 204

Sasha Aravkin, University of Washington .
An Interior Point Approach for Data Science

Abstract: Many important applications can be formulated as large-scale optimization problems, including classification in machine learning, data assimilation in weather prediction, inverse problems, and medical and seismic imaging. While first-order methods have proven widely successful in recent years, recent developments suggest that matrix-free second-order methods, such as interior-point methods, can be competitive.

We first develop a modeling framework for a wide range of problems, and show how conjugate representations can be exploited to design an interior point approach for this class. We then show several applications, with emphasis on modeling and problem structure, and end by discussing extensions to large-scale problems.