Correlation clustering of graphs and integers

S. Akiyama, L. Aszalós, L. Hajdu, A. Pethő

Research output: Contribution to journalArticle

Abstract

Correlation clustering can he modeled in ihe following way. Let A be a nonempty set, and ∼ be a symmetric binary relation on A. Consider a partition (clustering) P of A. We say that two distinct elements a, b ε A are in conflict, if a∼b, but a and b belong to different classes (clusters) of P, or if a ∼ b, however, these elements belong to the same class of P. The main objective in correlation clustering is to find an optimal P with respect to ∼, i.e. a clustering yielding the minimal number of conflicts. We note that correlation clustering, among others, plays an important role in machine learning. In this paper we provide results in three different, but closely connected directions. First we prove general new results for correlation clustering, using an alternative graph model of the problem. Then we deal with the correlation clustering of positive integers, with respect to a relation ∼ based on coprimality. Note that this part is in fact a survey of our earlier results. Finally, we consider the set of so-called iS'-units, which are positive integers having all prime divisors in a fixed finite set. Here we prove new results, again with respect to a relation defined by the help of coprimality. We note that interestingly, the shape of the optimal clustering radically differs for integers and, S-units.

Original languageEnglish
Pages (from-to)3-12
Number of pages10
JournalInfocommunications Journal
Volume6
Issue number4
Publication statusPublished - Dec 1 2014

Fingerprint

Learning systems

Keywords

  • Correlation clustering
  • Graphs
  • Integers
  • S-units

ASJC Scopus subject areas

  • Computer Science(all)
  • Electrical and Electronic Engineering

Cite this

Akiyama, S., Aszalós, L., Hajdu, L., & Pethő, A. (2014). Correlation clustering of graphs and integers. Infocommunications Journal, 6(4), 3-12.

Correlation clustering of graphs and integers. / Akiyama, S.; Aszalós, L.; Hajdu, L.; Pethő, A.

In: Infocommunications Journal, Vol. 6, No. 4, 01.12.2014, p. 3-12.

Research output: Contribution to journalArticle

Akiyama, S, Aszalós, L, Hajdu, L & Pethő, A 2014, 'Correlation clustering of graphs and integers', Infocommunications Journal, vol. 6, no. 4, pp. 3-12.
Akiyama S, Aszalós L, Hajdu L, Pethő A. Correlation clustering of graphs and integers. Infocommunications Journal. 2014 Dec 1;6(4):3-12.
Akiyama, S. ; Aszalós, L. ; Hajdu, L. ; Pethő, A. / Correlation clustering of graphs and integers. In: Infocommunications Journal. 2014 ; Vol. 6, No. 4. pp. 3-12.
@article{449851874e904aafa66e311b941422d1,
title = "Correlation clustering of graphs and integers",
abstract = "Correlation clustering can he modeled in ihe following way. Let A be a nonempty set, and ∼ be a symmetric binary relation on A. Consider a partition (clustering) P of A. We say that two distinct elements a, b ε A are in conflict, if a∼b, but a and b belong to different classes (clusters) of P, or if a ∼ b, however, these elements belong to the same class of P. The main objective in correlation clustering is to find an optimal P with respect to ∼, i.e. a clustering yielding the minimal number of conflicts. We note that correlation clustering, among others, plays an important role in machine learning. In this paper we provide results in three different, but closely connected directions. First we prove general new results for correlation clustering, using an alternative graph model of the problem. Then we deal with the correlation clustering of positive integers, with respect to a relation ∼ based on coprimality. Note that this part is in fact a survey of our earlier results. Finally, we consider the set of so-called iS'-units, which are positive integers having all prime divisors in a fixed finite set. Here we prove new results, again with respect to a relation defined by the help of coprimality. We note that interestingly, the shape of the optimal clustering radically differs for integers and, S-units.",
keywords = "Correlation clustering, Graphs, Integers, S-units",
author = "S. Akiyama and L. Aszal{\'o}s and L. Hajdu and A. Pethő",
year = "2014",
month = "12",
day = "1",
language = "English",
volume = "6",
pages = "3--12",
journal = "Infocommunications Journal",
issn = "2061-2079",
publisher = "Scientific Association for Infocommunications",
number = "4",

}

TY - JOUR

T1 - Correlation clustering of graphs and integers

AU - Akiyama, S.

AU - Aszalós, L.

AU - Hajdu, L.

AU - Pethő, A.

PY - 2014/12/1

Y1 - 2014/12/1

N2 - Correlation clustering can he modeled in ihe following way. Let A be a nonempty set, and ∼ be a symmetric binary relation on A. Consider a partition (clustering) P of A. We say that two distinct elements a, b ε A are in conflict, if a∼b, but a and b belong to different classes (clusters) of P, or if a ∼ b, however, these elements belong to the same class of P. The main objective in correlation clustering is to find an optimal P with respect to ∼, i.e. a clustering yielding the minimal number of conflicts. We note that correlation clustering, among others, plays an important role in machine learning. In this paper we provide results in three different, but closely connected directions. First we prove general new results for correlation clustering, using an alternative graph model of the problem. Then we deal with the correlation clustering of positive integers, with respect to a relation ∼ based on coprimality. Note that this part is in fact a survey of our earlier results. Finally, we consider the set of so-called iS'-units, which are positive integers having all prime divisors in a fixed finite set. Here we prove new results, again with respect to a relation defined by the help of coprimality. We note that interestingly, the shape of the optimal clustering radically differs for integers and, S-units.

AB - Correlation clustering can he modeled in ihe following way. Let A be a nonempty set, and ∼ be a symmetric binary relation on A. Consider a partition (clustering) P of A. We say that two distinct elements a, b ε A are in conflict, if a∼b, but a and b belong to different classes (clusters) of P, or if a ∼ b, however, these elements belong to the same class of P. The main objective in correlation clustering is to find an optimal P with respect to ∼, i.e. a clustering yielding the minimal number of conflicts. We note that correlation clustering, among others, plays an important role in machine learning. In this paper we provide results in three different, but closely connected directions. First we prove general new results for correlation clustering, using an alternative graph model of the problem. Then we deal with the correlation clustering of positive integers, with respect to a relation ∼ based on coprimality. Note that this part is in fact a survey of our earlier results. Finally, we consider the set of so-called iS'-units, which are positive integers having all prime divisors in a fixed finite set. Here we prove new results, again with respect to a relation defined by the help of coprimality. We note that interestingly, the shape of the optimal clustering radically differs for integers and, S-units.

KW - Correlation clustering

KW - Graphs

KW - Integers

KW - S-units

UR - http://www.scopus.com/inward/record.url?scp=84943603081&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84943603081&partnerID=8YFLogxK

M3 - Article

VL - 6

SP - 3

EP - 12

JO - Infocommunications Journal

JF - Infocommunications Journal

SN - 2061-2079

IS - 4

ER -