[Ml-stat-talks] ORFE Seminar: Mengdi Wang, February 14 at 4:30pm, Sherrerd Hall 101

Philippe Rigollet rigollet at Princeton.EDU
Mon Feb 11 13:47:39 EST 2013


=== ORFE Seminar Announcement === 

DATE:   Thursday, February 14, 2013   

TIME:   4:30pm 

LOCATION:   Room 101, Sherrerd Hall 

SPEAKER:  Mengdi Wang, MIT

TITLE:  Random Incremental Methods for Large-Scale Convex Optimization and Variational Inequalities

ABSTRACT:  
We consider stochastic methods for large-scale convex optimization and variational inequalities. We focus on special structures that lend themselves to sampling, such as when the constraint is the intersection of a large number of simpler sets, and/or the objective function is an expected value or is the sum of a large number of component functions. We propose new methods that combine elements of incremental constraint projection, stochastic gradient descent and proximal algorithm. These methods can be applied with various sampling schemes that are suitable for applications involving distributed implementation, large data set, or online learning. We use a unified framework to analyze the convergence and the rate of convergence of these methods, and show that the convergence involves an interplay between two coupled processes with different rates of convergence.




More information about the Ml-stat-talks mailing list