An Optimization Perspective on Data Analysis

Thursday, September 15, 2016 - 11:10am - 12:00pm
Keller 3-180
Moses Charikar (Stanford University)
In this talk, I will survey various data analysis questions cast as optimization problems. In many interesting cases, exact optima are hard to compute, and we turn to approximation guarantees. My talk will focus on clustering formulations, tensor decomposition and data analysis with outliers. I will highlight two cross cutting themes: (1) The use of convex relaxations that has proved to be a valuable tool in the design of algorithms with provable guarantees, and (2) Performance guarantees that go beyond worst case analysis.