Sharp kernel clustering algorithms and their associated Grothendieck inequalities

Subhash Khot, Assaf Naor

Research output: Contribution to journalArticle

Abstract

In the kernel clustering problem we are given a (large) n × n symmetric positive semidefinite matrix A = (aij) with ∑n i=1n j=1 aij=0 and a (small) k × k symmetric positive semidefinite matrix B = (bij). The goal is to find a partition {S1,...,Sk} of {1,...n} which maximizes ∑k i=1k i=1 (∑(p,q)εSi×Sj apq)bij. We design a polynomial time approximation algorithm that achieves an approximation ratio of R(B)2/C(B), where R(B) and C(B) are geometric parameters that depend only on the matrix B, defined as follows: if bij = 〈vi,vj〉 is the Gram matrix representation of B for some v1,....,vkεℝk then R(B) is the minimum radius of a Euclidean ball containing the points {v1,...,vk}. The parameter C(B) is defined as the maximum over all measurable partitions {A1,...,Ak} of ℝk-1 of the quantity ∑k i=1k j=1nij 〈Zi, Zj〉where for iε{1,...,k} the vector Zi ε ℝk-1 is the Gaussian moment of Ai. We also show that for every ε > 0, achieving an approximation guarantee of is Unique Games hard.

Original languageEnglish (US)
Pages (from-to)269-300
Number of pages32
JournalRandom Structures and Algorithms
Volume42
Issue number3
DOIs
StatePublished - May 2013

Fingerprint

Positive Semidefinite Matrix
Clustering algorithms
Clustering Algorithm
Partition
kernel
Gram Matrix
Matrix Representation
Approximation
Polynomial-time Algorithm
Approximation Algorithms
Euclidean
Ball
Maximise
Radius
Clustering
Game
Moment
Approximation algorithms
Polynomials
Design

Keywords

  • Approximation algorithms
  • Semidefinite programming
  • Unique Games hardness

ASJC Scopus subject areas

  • Computer Graphics and Computer-Aided Design
  • Software
  • Mathematics(all)
  • Applied Mathematics

Cite this

Sharp kernel clustering algorithms and their associated Grothendieck inequalities. / Khot, Subhash; Naor, Assaf.

In: Random Structures and Algorithms, Vol. 42, No. 3, 05.2013, p. 269-300.

Research output: Contribution to journalArticle

@article{661597498bca4c28a7c7d24b8b95337a,
title = "Sharp kernel clustering algorithms and their associated Grothendieck inequalities",
abstract = "In the kernel clustering problem we are given a (large) n × n symmetric positive semidefinite matrix A = (aij) with ∑n i=1 ∑n j=1 aij=0 and a (small) k × k symmetric positive semidefinite matrix B = (bij). The goal is to find a partition {S1,...,Sk} of {1,...n} which maximizes ∑k i=1 ∑k i=1 (∑(p,q)εSi×Sj apq)bij. We design a polynomial time approximation algorithm that achieves an approximation ratio of R(B)2/C(B), where R(B) and C(B) are geometric parameters that depend only on the matrix B, defined as follows: if bij = 〈vi,vj〉 is the Gram matrix representation of B for some v1,....,vkεℝk then R(B) is the minimum radius of a Euclidean ball containing the points {v1,...,vk}. The parameter C(B) is defined as the maximum over all measurable partitions {A1,...,Ak} of ℝk-1 of the quantity ∑k i=1 ∑k j=1nij 〈Zi, Zj〉where for iε{1,...,k} the vector Zi ε ℝk-1 is the Gaussian moment of Ai. We also show that for every ε > 0, achieving an approximation guarantee of is Unique Games hard.",
keywords = "Approximation algorithms, Semidefinite programming, Unique Games hardness",
author = "Subhash Khot and Assaf Naor",
year = "2013",
month = "5",
doi = "10.1002/rsa.20398",
language = "English (US)",
volume = "42",
pages = "269--300",
journal = "Random Structures and Algorithms",
issn = "1042-9832",
publisher = "John Wiley and Sons Ltd",
number = "3",

}

TY - JOUR

T1 - Sharp kernel clustering algorithms and their associated Grothendieck inequalities

AU - Khot, Subhash

AU - Naor, Assaf

PY - 2013/5

Y1 - 2013/5

N2 - In the kernel clustering problem we are given a (large) n × n symmetric positive semidefinite matrix A = (aij) with ∑n i=1 ∑n j=1 aij=0 and a (small) k × k symmetric positive semidefinite matrix B = (bij). The goal is to find a partition {S1,...,Sk} of {1,...n} which maximizes ∑k i=1 ∑k i=1 (∑(p,q)εSi×Sj apq)bij. We design a polynomial time approximation algorithm that achieves an approximation ratio of R(B)2/C(B), where R(B) and C(B) are geometric parameters that depend only on the matrix B, defined as follows: if bij = 〈vi,vj〉 is the Gram matrix representation of B for some v1,....,vkεℝk then R(B) is the minimum radius of a Euclidean ball containing the points {v1,...,vk}. The parameter C(B) is defined as the maximum over all measurable partitions {A1,...,Ak} of ℝk-1 of the quantity ∑k i=1 ∑k j=1nij 〈Zi, Zj〉where for iε{1,...,k} the vector Zi ε ℝk-1 is the Gaussian moment of Ai. We also show that for every ε > 0, achieving an approximation guarantee of is Unique Games hard.

AB - In the kernel clustering problem we are given a (large) n × n symmetric positive semidefinite matrix A = (aij) with ∑n i=1 ∑n j=1 aij=0 and a (small) k × k symmetric positive semidefinite matrix B = (bij). The goal is to find a partition {S1,...,Sk} of {1,...n} which maximizes ∑k i=1 ∑k i=1 (∑(p,q)εSi×Sj apq)bij. We design a polynomial time approximation algorithm that achieves an approximation ratio of R(B)2/C(B), where R(B) and C(B) are geometric parameters that depend only on the matrix B, defined as follows: if bij = 〈vi,vj〉 is the Gram matrix representation of B for some v1,....,vkεℝk then R(B) is the minimum radius of a Euclidean ball containing the points {v1,...,vk}. The parameter C(B) is defined as the maximum over all measurable partitions {A1,...,Ak} of ℝk-1 of the quantity ∑k i=1 ∑k j=1nij 〈Zi, Zj〉where for iε{1,...,k} the vector Zi ε ℝk-1 is the Gaussian moment of Ai. We also show that for every ε > 0, achieving an approximation guarantee of is Unique Games hard.

KW - Approximation algorithms

KW - Semidefinite programming

KW - Unique Games hardness

UR - http://www.scopus.com/inward/record.url?scp=84875547836&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84875547836&partnerID=8YFLogxK

U2 - 10.1002/rsa.20398

DO - 10.1002/rsa.20398

M3 - Article

VL - 42

SP - 269

EP - 300

JO - Random Structures and Algorithms

JF - Random Structures and Algorithms

SN - 1042-9832

IS - 3

ER -