Nonasymptotic convergence rates for cooperative learning over time-varying directed graphs

Angelia Nedich, Alex Olshevsky, Cesar A. Uribe

Research output: Chapter in Book/Report/Conference proceedingConference contribution

23 Citations (Scopus)

Abstract

We study the problem of cooperative learning with a network of agents where some agents repeatedly access information about a random variable with unknown distribution. The group objective is to globally agree on a joint hypothesis (distribution) that best describes the observed data at all nodes. The agents interact with their neighbors in an unknown sequence of time-varying directed graphs. Following the pioneering work of Jadbabaie, Molavi, Sandroni, and Tahbaz-Salehi and others, we propose local learning dynamics which combine Bayesian updates at each node with a local aggregation rule of private agent signals. We show that these learning dynamics drive all agents to the set of hypotheses which best explain the data collected at all nodes as long as the sequence of interconnection graphs is uniformly strongly connected. Our main result establishes a non-asymptotic, explicit, geometric convergence rate for the learning dynamic.

Original languageEnglish (US)
Title of host publicationACC 2015 - 2015 American Control Conference
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages5884-5889
Number of pages6
Volume2015-July
ISBN (Electronic)9781479986842
DOIs
StatePublished - Jul 28 2015
Externally publishedYes
Event2015 American Control Conference, ACC 2015 - Chicago, United States
Duration: Jul 1 2015Jul 3 2015

Other

Other2015 American Control Conference, ACC 2015
CountryUnited States
CityChicago
Period7/1/157/3/15

Fingerprint

Directed graphs
Random variables
Agglomeration

ASJC Scopus subject areas

  • Electrical and Electronic Engineering

Cite this

Nedich, A., Olshevsky, A., & Uribe, C. A. (2015). Nonasymptotic convergence rates for cooperative learning over time-varying directed graphs. In ACC 2015 - 2015 American Control Conference (Vol. 2015-July, pp. 5884-5889). [7172262] Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ACC.2015.7172262

Nonasymptotic convergence rates for cooperative learning over time-varying directed graphs. / Nedich, Angelia; Olshevsky, Alex; Uribe, Cesar A.

ACC 2015 - 2015 American Control Conference. Vol. 2015-July Institute of Electrical and Electronics Engineers Inc., 2015. p. 5884-5889 7172262.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Nedich, A, Olshevsky, A & Uribe, CA 2015, Nonasymptotic convergence rates for cooperative learning over time-varying directed graphs. in ACC 2015 - 2015 American Control Conference. vol. 2015-July, 7172262, Institute of Electrical and Electronics Engineers Inc., pp. 5884-5889, 2015 American Control Conference, ACC 2015, Chicago, United States, 7/1/15. https://doi.org/10.1109/ACC.2015.7172262
Nedich A, Olshevsky A, Uribe CA. Nonasymptotic convergence rates for cooperative learning over time-varying directed graphs. In ACC 2015 - 2015 American Control Conference. Vol. 2015-July. Institute of Electrical and Electronics Engineers Inc. 2015. p. 5884-5889. 7172262 https://doi.org/10.1109/ACC.2015.7172262
Nedich, Angelia ; Olshevsky, Alex ; Uribe, Cesar A. / Nonasymptotic convergence rates for cooperative learning over time-varying directed graphs. ACC 2015 - 2015 American Control Conference. Vol. 2015-July Institute of Electrical and Electronics Engineers Inc., 2015. pp. 5884-5889
@inproceedings{5fb4fe3a22f5424ab1f2f55e220ce2ab,
title = "Nonasymptotic convergence rates for cooperative learning over time-varying directed graphs",
abstract = "We study the problem of cooperative learning with a network of agents where some agents repeatedly access information about a random variable with unknown distribution. The group objective is to globally agree on a joint hypothesis (distribution) that best describes the observed data at all nodes. The agents interact with their neighbors in an unknown sequence of time-varying directed graphs. Following the pioneering work of Jadbabaie, Molavi, Sandroni, and Tahbaz-Salehi and others, we propose local learning dynamics which combine Bayesian updates at each node with a local aggregation rule of private agent signals. We show that these learning dynamics drive all agents to the set of hypotheses which best explain the data collected at all nodes as long as the sequence of interconnection graphs is uniformly strongly connected. Our main result establishes a non-asymptotic, explicit, geometric convergence rate for the learning dynamic.",
author = "Angelia Nedich and Alex Olshevsky and Uribe, {Cesar A.}",
year = "2015",
month = "7",
day = "28",
doi = "10.1109/ACC.2015.7172262",
language = "English (US)",
volume = "2015-July",
pages = "5884--5889",
booktitle = "ACC 2015 - 2015 American Control Conference",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
address = "United States",

}

TY - GEN

T1 - Nonasymptotic convergence rates for cooperative learning over time-varying directed graphs

AU - Nedich, Angelia

AU - Olshevsky, Alex

AU - Uribe, Cesar A.

PY - 2015/7/28

Y1 - 2015/7/28

N2 - We study the problem of cooperative learning with a network of agents where some agents repeatedly access information about a random variable with unknown distribution. The group objective is to globally agree on a joint hypothesis (distribution) that best describes the observed data at all nodes. The agents interact with their neighbors in an unknown sequence of time-varying directed graphs. Following the pioneering work of Jadbabaie, Molavi, Sandroni, and Tahbaz-Salehi and others, we propose local learning dynamics which combine Bayesian updates at each node with a local aggregation rule of private agent signals. We show that these learning dynamics drive all agents to the set of hypotheses which best explain the data collected at all nodes as long as the sequence of interconnection graphs is uniformly strongly connected. Our main result establishes a non-asymptotic, explicit, geometric convergence rate for the learning dynamic.

AB - We study the problem of cooperative learning with a network of agents where some agents repeatedly access information about a random variable with unknown distribution. The group objective is to globally agree on a joint hypothesis (distribution) that best describes the observed data at all nodes. The agents interact with their neighbors in an unknown sequence of time-varying directed graphs. Following the pioneering work of Jadbabaie, Molavi, Sandroni, and Tahbaz-Salehi and others, we propose local learning dynamics which combine Bayesian updates at each node with a local aggregation rule of private agent signals. We show that these learning dynamics drive all agents to the set of hypotheses which best explain the data collected at all nodes as long as the sequence of interconnection graphs is uniformly strongly connected. Our main result establishes a non-asymptotic, explicit, geometric convergence rate for the learning dynamic.

UR - http://www.scopus.com/inward/record.url?scp=84940917154&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84940917154&partnerID=8YFLogxK

U2 - 10.1109/ACC.2015.7172262

DO - 10.1109/ACC.2015.7172262

M3 - Conference contribution

AN - SCOPUS:84940917154

VL - 2015-July

SP - 5884

EP - 5889

BT - ACC 2015 - 2015 American Control Conference

PB - Institute of Electrical and Electronics Engineers Inc.

ER -