Parallel Successive Convex Approximation for Nonsmooth Nonconvex Optimization

Thumbnail Image
Supplemental Files
Date
2014-01-01
Authors
Razaviyayn, Mesiam
Hong, Mingyi
Luo, Zhi-Quan
Pang, Jong-Shi
Major Professor
Advisor
Committee Member
Journal Title
Journal ISSN
Volume Title
Publisher
Authors
Person
Hong, Mingyi
Assistant Professor
Research Projects
Organizational Units
Organizational Unit
Industrial and Manufacturing Systems Engineering
The Department of Industrial and Manufacturing Systems Engineering teaches the design, analysis, and improvement of the systems and processes in manufacturing, consulting, and service industries by application of the principles of engineering. The Department of General Engineering was formed in 1929. In 1956 its name changed to Department of Industrial Engineering. In 1989 its name changed to the Department of Industrial and Manufacturing Systems Engineering.
Journal Issue
Is Version Of
Versions
Series
Department
Industrial and Manufacturing Systems Engineering
Abstract

Consider the problem of minimizing the sum of a smooth (possibly non-convex) and a convex (possibly nonsmooth) function involving a large number of variables. A popular approach to solve this problem is the block coordinate descent (BCD) method whereby at each iteration only one variable block is updated while the remaining variables are held fixed. With the recent advances in the developments of the multi-core parallel processing technology, it is desirable to parallelize the BCD method by allowing multiple blocks to be updated simultaneously at each iteration of the algorithm. In this work, we propose an inexact parallel BCD approach where at each iteration, a subset of the variables is updated in parallel by minimizing convex approximations of the original objective function. We investigate the convergence of this parallel BCD method for both randomized and cyclic variable selection rules. We analyze the asymptotic and non-asymptotic convergence behavior of the algorithm for both convex and non-convex objective functions. The numerical experiments suggest that for a special case of Lasso minimization problem, the cyclic block selection rule can outperform the randomized rule.

Comments

This is a proceeding from the 28th Conference on Neural Information Processing Systems (2014). Posted with permission.

Description
Keywords
Citation
DOI
Source
Copyright
Wed Jan 01 00:00:00 UTC 2014