DocumentCode :
2856600
Title :
Accelerated dual descent for network optimization
Author :
Zargham, M. ; Ribeiro, A. ; Ozdaglar, A. ; Jadbabaie, A.
Author_Institution :
Dept. of Electr. & Syst. Eng., Univ. of Pennsylvania, Philadelphia, PA, USA
fYear :
2011
fDate :
June 29 2011-July 1 2011
Firstpage :
2663
Lastpage :
2668
Abstract :
Dual descent methods are commonly used to solve network optimization problems because their implementation can be distributed through the network. However, their convergence rates are typically very slow. This paper introduces a family of dual descent algorithms that use approximate Newton directions to accelerate the convergence rate of conventional dual descent. These approximate directions can be computed using local information exchanges thereby retaining the benefits of distributed implementations. The approximate Newton directions are obtained through matrix splitting techniques and sparse Taylor approximations of the inverse Hessian. We show that, similarly to conventional Newton methods, the proposed algorithm exhibits superlinear convergence within a neighborhood of the optimal value. Numerical analysis corroborates that convergence times are between one to two orders of magnitude faster than existing distributed optimization methods.
Keywords :
Newton method; approximation theory; convergence of numerical methods; distributed algorithms; network theory (graphs); optimisation; sparse matrices; accelerated dual descent method; approximate Newton directions; inverse Hessian; local information exchange; matrix splitting technique; network optimization problem; numerical analysis; sparse Taylor approximation; Approximation algorithms; Approximation methods; Convergence; Laplace equations; Newton method; Optimization; Taylor series;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
American Control Conference (ACC), 2011
Conference_Location :
San Francisco, CA
ISSN :
0743-1619
Print_ISBN :
978-1-4577-0080-4
Type :
conf
DOI :
10.1109/ACC.2011.5991367
Filename :
5991367
Link To Document :
بازگشت