ANC Workshop Matt Graham/Zhanxing Zhu Chair: Charles Sutton
What 


When 
Jun 16, 2015 from 11:00 AM to 12:00 PM 
Where  IF Room 4.31/4.33 
Add event to calendar 
vCal iCal 
Matt Graham
Title: Dynamics, geometry and randomness: sampling from Boltzmann machine relaxations
Zhanxing Zhu
Title: Stochastic Parallel Block Coordinate Descent for Separable Saddle Point Problems
We consider a general convexconcave saddle point problem with a separable structure, especially for nonstrongly convex functions. Given this form of problem, we propose a simple and efficient stochastic block coordinate descent method using adaptive primaldual updates, which enables flexible parallel optimization for largescale optimization. Our method shares the efficiency and flexibility of coordinate descent methods while keeping the simplicity of primaldual methods and utilizing the structure of the separable convexconcave saddle point problem.
The method is capable of solving a wide range of machine learning applications, including robust principal component analysis, Lasso and feature selection by group Lasso, etc. Both theoretically and empirically, I will show that the proposed method performs significantly better than other stateoftheart methods in all these applications.