DrSVM

Distributed random projection algorithms for SVMs

Soomin Lee, Angelia Nedich

Research output: Contribution to journalArticle

4 Citations (Scopus)

Abstract

We present distributed random projected gradient algorithms for Support Vector Machines (SVMs) that can be used by multiple agents connected over a time-varying network. The goal is for the agents to cooperatively find the same maximum margin hyperplane. In the primal SVM formulation, the objective function can be represented as a sum of convex functions and the constraint set is an intersection of multiple halfspaces. Each agent minimizes a local objective subject to a local constraint set. It maintains its own estimate sequence and communicates with its neighbors. More specifically, each agent calculates weighted averages of the received estimates and its own estimate, adjust the estimate by using gradient information of its local objective function and project onto a subset of its local constraint set. At each iteration, an agent considers only one halfspace since projection onto a single halfspace is easy. We also consider the convergence behavior of the algorithms and prove that all the estimates of agents converge to the same limit point in the optimal set.

Original languageEnglish (US)
Article number6425875
Pages (from-to)5286-5291
Number of pages6
JournalUnknown Journal
DOIs
StatePublished - 2012
Externally publishedYes

Fingerprint

Random Projection
Projection Algorithm
Support vector machines
Support Vector Machine
Half-space
Estimate
Time varying networks
Objective function
Projected Gradient
Set theory
Gradient Algorithm
Limit Point
Weighted Average
Hyperplane
Margin
Convex function
Time-varying
Intersection
Projection
Gradient

ASJC Scopus subject areas

  • Control and Systems Engineering
  • Modeling and Simulation
  • Control and Optimization

Cite this

DrSVM : Distributed random projection algorithms for SVMs. / Lee, Soomin; Nedich, Angelia.

In: Unknown Journal, 2012, p. 5286-5291.

Research output: Contribution to journalArticle

@article{bf7fd669c8af473fb48cb12eb40ce1d5,
title = "DrSVM: Distributed random projection algorithms for SVMs",
abstract = "We present distributed random projected gradient algorithms for Support Vector Machines (SVMs) that can be used by multiple agents connected over a time-varying network. The goal is for the agents to cooperatively find the same maximum margin hyperplane. In the primal SVM formulation, the objective function can be represented as a sum of convex functions and the constraint set is an intersection of multiple halfspaces. Each agent minimizes a local objective subject to a local constraint set. It maintains its own estimate sequence and communicates with its neighbors. More specifically, each agent calculates weighted averages of the received estimates and its own estimate, adjust the estimate by using gradient information of its local objective function and project onto a subset of its local constraint set. At each iteration, an agent considers only one halfspace since projection onto a single halfspace is easy. We also consider the convergence behavior of the algorithms and prove that all the estimates of agents converge to the same limit point in the optimal set.",
author = "Soomin Lee and Angelia Nedich",
year = "2012",
doi = "10.1109/CDC.2012.6425875",
language = "English (US)",
pages = "5286--5291",
journal = "Scanning Electron Microscopy",
issn = "0586-5581",
publisher = "Scanning Microscopy International",

}

TY - JOUR

T1 - DrSVM

T2 - Distributed random projection algorithms for SVMs

AU - Lee, Soomin

AU - Nedich, Angelia

PY - 2012

Y1 - 2012

N2 - We present distributed random projected gradient algorithms for Support Vector Machines (SVMs) that can be used by multiple agents connected over a time-varying network. The goal is for the agents to cooperatively find the same maximum margin hyperplane. In the primal SVM formulation, the objective function can be represented as a sum of convex functions and the constraint set is an intersection of multiple halfspaces. Each agent minimizes a local objective subject to a local constraint set. It maintains its own estimate sequence and communicates with its neighbors. More specifically, each agent calculates weighted averages of the received estimates and its own estimate, adjust the estimate by using gradient information of its local objective function and project onto a subset of its local constraint set. At each iteration, an agent considers only one halfspace since projection onto a single halfspace is easy. We also consider the convergence behavior of the algorithms and prove that all the estimates of agents converge to the same limit point in the optimal set.

AB - We present distributed random projected gradient algorithms for Support Vector Machines (SVMs) that can be used by multiple agents connected over a time-varying network. The goal is for the agents to cooperatively find the same maximum margin hyperplane. In the primal SVM formulation, the objective function can be represented as a sum of convex functions and the constraint set is an intersection of multiple halfspaces. Each agent minimizes a local objective subject to a local constraint set. It maintains its own estimate sequence and communicates with its neighbors. More specifically, each agent calculates weighted averages of the received estimates and its own estimate, adjust the estimate by using gradient information of its local objective function and project onto a subset of its local constraint set. At each iteration, an agent considers only one halfspace since projection onto a single halfspace is easy. We also consider the convergence behavior of the algorithms and prove that all the estimates of agents converge to the same limit point in the optimal set.

UR - http://www.scopus.com/inward/record.url?scp=84874236050&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84874236050&partnerID=8YFLogxK

U2 - 10.1109/CDC.2012.6425875

DO - 10.1109/CDC.2012.6425875

M3 - Article

SP - 5286

EP - 5291

JO - Scanning Electron Microscopy

JF - Scanning Electron Microscopy

SN - 0586-5581

M1 - 6425875

ER -