[Home]
[
News]
[ Events]
[ People]
[ Research]
[ Education]
[Visitor Info]
[UCSD Only]
[Admin]
Fast Alternating Direction Methods for Optimization
Tom Goldstein
Post-Doctoral Fellow at the Rice University Department of Electrical Engineering
Abstract:
Alternating direction methods are a commonplace tool for general mathematical programming and optimization. These methods have become particularly important in the field of variational image processing, which frequently requires the minimization of non-differentiable objectives. This paper considers accelerated (i.e. fast) variants of two common alternating direction methods: the Alternating Direction Method of Multipliers (ADMM) and the Alternating Minimization Algorithm (AMA). The proposed acceleration is of the form first proposed by Nesterov for gradient descent methods. In the case that the objective function is strongly convex, global convergence bounds are provided for both classical and accelerated variants of the methods. Numerical examples are presented to demonstrating the superior performance of the fast methods.
Thursday, November 15, 2012
11:00AM AP&M 2402
News]
[ Events]
[ People]
[ Research]
[ Education]
[Visitor Info]
[UCSD Only]
[Admin]
Directors:
Randolph E. Bank
Philip E. Gill
Michael Holst
Administrative Contact:
[ Click to Send Email ]
Randolph E. Bank
Philip E. Gill
Michael Holst
Administrative Contact:
[ Click to Send Email ]
Tom Goldstein
Post-Doctoral Fellow at the Rice University Department of Electrical Engineering
Abstract:
Alternating direction methods are a commonplace tool for general mathematical programming and optimization. These methods have become particularly important in the field of variational image processing, which frequently requires the minimization of non-differentiable objectives. This paper considers accelerated (i.e. fast) variants of two common alternating direction methods: the Alternating Direction Method of Multipliers (ADMM) and the Alternating Minimization Algorithm (AMA). The proposed acceleration is of the form first proposed by Nesterov for gradient descent methods. In the case that the objective function is strongly convex, global convergence bounds are provided for both classical and accelerated variants of the methods. Numerical examples are presented to demonstrating the superior performance of the fast methods.
Thursday, November 15, 2012
11:00AM AP&M 2402