Gossip algorithms for clustering problems

Linh Thi Hoai Nguyen, Takayuki Wada, Izumi Masubuchi, Toru Asai, Yasumasa Fujisaki

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

1 Citation (Scopus)

Abstract

This paper presents several algorithms for multi-dimensional data clustering. The main difference of the algorithms introduced in this paper comparing to known ones is that the algorithms are multi-agent based and gossip. Hence, the number of computation in each step is not too large even for large data and the computational burden can be distributed. The basic algorithm is described as follows. Each data object is assigned to an agent in a network and its attributes constitute to a vector which is considered as the initial state of the agent. A common confidence threshold is set for all of the agents. The states of agents in the network will be updated time by time according to an iterative procedure: At each time, (i) two arbitrary agents are chosen randomly, (ii) they exchange their states, and (iii) if the distance between their states does not exceed the confidence threshold, they update their states as the average of the two. Under a certain condition, this algorithm converges almost surely in an equilibrium point such that any two agents either have the same final state or have distinct states which differ more than the confidence threshold from each other. In this way, the given data objects are partitioned into groups such that ones in the same group correspond to the agents with the same final state. Some modified algorithms for improving the performance of clustering and/or getting some prescribed condition are also introduced.

Original languageEnglish
Title of host publicationIECON 2015 - 41st Annual Conference of the IEEE Industrial Electronics Society
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages589-594
Number of pages6
ISBN (Electronic)9781479917624
DOIs
Publication statusPublished - 2015
Externally publishedYes
Event41st Annual Conference of the IEEE Industrial Electronics Society, IECON 2015 - Yokohama, Japan
Duration: Nov 9 2015Nov 12 2015

Publication series

NameIECON 2015 - 41st Annual Conference of the IEEE Industrial Electronics Society

Other

Other41st Annual Conference of the IEEE Industrial Electronics Society, IECON 2015
CountryJapan
CityYokohama
Period11/9/1511/12/15

All Science Journal Classification (ASJC) codes

  • Electrical and Electronic Engineering
  • Industrial and Manufacturing Engineering

Cite this