DocumentCode :
730453
Title :
An approximate Newton method for distributed optimization
Author :
Mokhtari, Aryan ; Qing Ling ; Ribeiro, Alejandro
Author_Institution :
Dept. of Electr. & Syst. Eng., Univ. of Pennsylvania, Philadelphia, PA, USA
fYear :
2015
fDate :
19-24 April 2015
Firstpage :
2959
Lastpage :
2963
Abstract :
Agents of a network have access to strongly convex local functions fi and attempt to minimize the aggregate function f(x) = Σi=1nfi(x) while relying on variable exchanges with neighboring nodes. Various methods to solve this distributed optimization problem exist but they all rely on first order information. This paper introduces Network Newton, a method that incorporates second order information via distributed evaluation of approximations to Newton steps. The method is shown to converge linearly and to do so while exhibiting a quadratic phase. Numerical analyses show substantial reductions in convergence times relative to existing (first order) alternatives.
Keywords :
Newton method; approximation theory; convergence of numerical methods; convex programming; minimisation; aggregate function minimization; approximate network Newton method; convergence times; distributed optimization problem; multiagent network; neighboring nodes; numerical analysis; strongly convex local functions; variable exchanges; Approximation algorithms; Approximation methods; Artificial neural networks; Convergence; Linear programming; Nickel; Optimization; Multi-agent network; Newton method; distributed optimization;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Acoustics, Speech and Signal Processing (ICASSP), 2015 IEEE International Conference on
Conference_Location :
South Brisbane, QLD
Type :
conf
DOI :
10.1109/ICASSP.2015.7178513
Filename :
7178513
Link To Document :
بازگشت