Robust Regression and Efficient Optimization
Yaoliang Yu (University of Alberta)
NICTA SML SEMINARDATE: 2013-05-16
TIME: 11:15:00 - 12:15:00
LOCATION: NICTA - 7 London Circuit
CONTACT: JavaScript must be enabled to display this email address.
ABSTRACT:
Despite the variety of robust regression methods that have been developed, current approaches are either NP-hard, or allow unbounded response to even a single leverage point. In the first part of the talk, we present a general formulation for robust regressiona"Variational M-estimationa"that unifies a number of robust regression methods while allowing a tractable approximation strategy. We develop an estimator that requires only polynomial-time, while achieving certain robustness and consistency guarantees. In the second part of the talk we will present the generalized conditional gradient algorithm for minimizing a smooth loss function regularized by a nonsmooth term. We prove its convergence rate and show its promising performance on the matrix completion problem. Some recent extensions on how to efficiently compute the polar/proximal will also be discussed.
BIO:
Yaoliang Yu is a Ph.D student from University of Alberta, working under the supervision of Dale Schuurmans and Csaba Szepesvari. His main research interests include convex optimization, kernel methods, robust regression and nonparametric statistics.
