Jorge Cortés
Professor
Cymer Corporation Endowed Chair
Noise-to-state exponentially stable distributed convex optimization on
weight-balanced digraphs
D. Mateos-Nuñez, J. Cortés
SIAM Journal on Control and Optimization 54 (1) (2016), 266-290
Abstract
This paper studies the robustness properties against additive
persistent noise of a class of distributed continuous-time
algorithms for convex optimization. A team of agents, each with its
own private objective function, seeks to collectively determine the
global decision vector that minimizes the sum of the objective
functions. The team communicates over a weight-balanced, strongly
connected digraph and both inter-agent communication and agent
computation are corrupted by noise. Under the proposed distributed
algorithm, each agent updates its estimate of the global solution
using the gradient of its local objective function while, at the
same time, seeking to agree with its neighbors' estimates via
proportional-integral feedback on their disagreement. Under mild
conditions on the local objective functions, we show that this
strategy is noise-to-state exponentially stable in second moment
with respect to the optimal solution. Our technical approach
combines notions and tools from graph theory, stochastic
differential equations, Lyapunov stability analysis, and
co-coercivity of vector fields. Simulations illustrate our results.
pdf   |   ps.gz
Mechanical and Aerospace Engineering,
University of California, San Diego
9500 Gilman Dr,
La Jolla, California, 92093-0411
Ph: 1-858-822-7930
Fax: 1-858-822-3107
cortes at ucsd.edu
Skype id:
jorgilliyo