Joint Math-Stat Colloquium
Thursday, May 8th 2014 at 4:10pm, MSB 1147 (Colloquium Room)
Speaker: Laurent El Ghaoui, UC Berkeley
Title: "Robust Sketching for Large-Scale Optimization"
Abstract: In the recent years there has been a lot of interest in approximating a data matrix by a "sketch", that is, a simpler matrix that preserves some property of interest, and with which computations can be performed faster than the original. We consider the idea in the context of solving a linear or convex quadratic program, and develop the technique of "robust sketching", which entails replacing the coefficient matrix with a sketch, but keeping track of the error thus made, via robust optimization. We examine applications of the concept in the areas of sparse machine learning and in the context of very large LPs arising in energy management.