Degree Type

Dissertation

Date of Award

2017

Degree Name

Doctor of Philosophy

Department

Industrial and Manufacturing Systems Engineering

Major

Industrial and Manufacturing Systems Engineering

First Advisor

Mingyi Hong

Second Advisor

Gary Mirka

Abstract

This thesis addresses the problem of distributed optimization and learning over multi-agent networks. Our main focus is to design efficient algorithms for a class of nonconvex problems, defined over networks in which each agent/node only has partial knowledge about the entire problem. Multi-agent nonconvex optimization has gained much attention recently due to its wide applications in big data analysis, sensor networks, signal processing, multi-agent network, resource allocation, communication networks, just to name a few. In this work, we develop a general class of primal-dual algorithms for distributed optimization problems in challenging setups, such as nonconvexity in loss functions, nonsmooth regularizations, and coupling constraints. Further, we consider different setup where each agent can only access the zeroth-order information (i.e., the functional values) of its local functions. Rigorous convergence and rate of convergence analysis is provided for the proposed algorithms. Our work represents one of the first attempts to address nonconvex optimization and learning over networks.

DOI

https://doi.org/10.31274/etd-180810-5769

Copyright Owner

Davood Hajinezhad

Language

en

File Format

application/pdf

File Size

207 pages

Share

COinS