Stochastic Gradient-Push for Strongly Convex Functions on Time-Varying Directed Graphs

Angelia Nedich, Alex Olshevsky

Research output: Contribution to journalArticle

39 Citations (Scopus)

Abstract

We investigate the convergence rate of the recently proposed subgradient-push method for distributed optimization over time-varying directed graphs. The subgradient-push method can be implemented in a distributed way without requiring knowledge of either the number of agents or the graph sequence; each node is only required to know its out-degree at each time. Our main result is a convergence rate of O((ln t)/t) for strongly convex functions with Lipschitz gradients even if only stochastic gradient samples are available; this is asymptotically faster than the O((ln t)/t) rate previously known for (general) convex functions.

Original languageEnglish (US)
Article number7405263
Pages (from-to)3936-3947
Number of pages12
JournalIEEE Transactions on Automatic Control
Volume61
Issue number12
DOIs
StatePublished - Dec 1 2016
Externally publishedYes

Fingerprint

Directed graphs

Keywords

  • Distributed algorithms
  • gradient methods
  • parameter estimation

ASJC Scopus subject areas

  • Control and Systems Engineering
  • Computer Science Applications
  • Electrical and Electronic Engineering

Cite this

Stochastic Gradient-Push for Strongly Convex Functions on Time-Varying Directed Graphs. / Nedich, Angelia; Olshevsky, Alex.

In: IEEE Transactions on Automatic Control, Vol. 61, No. 12, 7405263, 01.12.2016, p. 3936-3947.

Research output: Contribution to journalArticle

@article{4eebc5719c8148459fc2ebaff4d710a4,
title = "Stochastic Gradient-Push for Strongly Convex Functions on Time-Varying Directed Graphs",
abstract = "We investigate the convergence rate of the recently proposed subgradient-push method for distributed optimization over time-varying directed graphs. The subgradient-push method can be implemented in a distributed way without requiring knowledge of either the number of agents or the graph sequence; each node is only required to know its out-degree at each time. Our main result is a convergence rate of O((ln t)/t) for strongly convex functions with Lipschitz gradients even if only stochastic gradient samples are available; this is asymptotically faster than the O((ln t)/t) rate previously known for (general) convex functions.",
keywords = "Distributed algorithms, gradient methods, parameter estimation",
author = "Angelia Nedich and Alex Olshevsky",
year = "2016",
month = "12",
day = "1",
doi = "10.1109/TAC.2016.2529285",
language = "English (US)",
volume = "61",
pages = "3936--3947",
journal = "IEEE Transactions on Automatic Control",
issn = "0018-9286",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
number = "12",

}

TY - JOUR

T1 - Stochastic Gradient-Push for Strongly Convex Functions on Time-Varying Directed Graphs

AU - Nedich, Angelia

AU - Olshevsky, Alex

PY - 2016/12/1

Y1 - 2016/12/1

N2 - We investigate the convergence rate of the recently proposed subgradient-push method for distributed optimization over time-varying directed graphs. The subgradient-push method can be implemented in a distributed way without requiring knowledge of either the number of agents or the graph sequence; each node is only required to know its out-degree at each time. Our main result is a convergence rate of O((ln t)/t) for strongly convex functions with Lipschitz gradients even if only stochastic gradient samples are available; this is asymptotically faster than the O((ln t)/t) rate previously known for (general) convex functions.

AB - We investigate the convergence rate of the recently proposed subgradient-push method for distributed optimization over time-varying directed graphs. The subgradient-push method can be implemented in a distributed way without requiring knowledge of either the number of agents or the graph sequence; each node is only required to know its out-degree at each time. Our main result is a convergence rate of O((ln t)/t) for strongly convex functions with Lipschitz gradients even if only stochastic gradient samples are available; this is asymptotically faster than the O((ln t)/t) rate previously known for (general) convex functions.

KW - Distributed algorithms

KW - gradient methods

KW - parameter estimation

UR - http://www.scopus.com/inward/record.url?scp=85003856207&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85003856207&partnerID=8YFLogxK

U2 - 10.1109/TAC.2016.2529285

DO - 10.1109/TAC.2016.2529285

M3 - Article

AN - SCOPUS:85003856207

VL - 61

SP - 3936

EP - 3947

JO - IEEE Transactions on Automatic Control

JF - IEEE Transactions on Automatic Control

SN - 0018-9286

IS - 12

M1 - 7405263

ER -