Instance-level constraint-based semisupervised learning with imposed space-partitioning

Research output: Contribution to journalArticle

2 Citations (Scopus)

Abstract

A new method for semisupervised learning from pairwise sample (must- and cannot-link) constraints is introduced. It addresses an important limitation of many existing methods, whose solutions do not achieve effective propagation of the constraint information to unconstrained samples. We overcome this limitation by constraining the solution to comport with a smooth (soft) class partition of the feature space, which necessarily entails constraint propagation and generalization to unconstrained samples. This is achieved via a parameterized mean-field approximation to the posterior distribution over component assignments, with the parameterization chosen to match the representation power of the chosen (generative) mixture density family. Unlike many existing methods, our method flexibly models classes using a variable number of components, which allows it to learn complex class boundaries. Also, unlike most of the methods, ours estimates the number of latent classes present in the data. Experiments on synthetic data and data sets from the UC Irvine machine learning repository show that, overall, our method achieves significant improvements in classification performance compared with the existing methods.

Original languageEnglish (US)
Article number6695785
Pages (from-to)1520-1537
Number of pages18
JournalIEEE Transactions on Neural Networks and Learning Systems
Volume25
Issue number8
DOIs
StatePublished - Jan 1 2014

Fingerprint

Parameterization
Learning systems
Experiments

All Science Journal Classification (ASJC) codes

  • Software
  • Computer Science Applications
  • Computer Networks and Communications
  • Artificial Intelligence

Cite this

@article{20f3405df30147de85499461c070c065,
title = "Instance-level constraint-based semisupervised learning with imposed space-partitioning",
abstract = "A new method for semisupervised learning from pairwise sample (must- and cannot-link) constraints is introduced. It addresses an important limitation of many existing methods, whose solutions do not achieve effective propagation of the constraint information to unconstrained samples. We overcome this limitation by constraining the solution to comport with a smooth (soft) class partition of the feature space, which necessarily entails constraint propagation and generalization to unconstrained samples. This is achieved via a parameterized mean-field approximation to the posterior distribution over component assignments, with the parameterization chosen to match the representation power of the chosen (generative) mixture density family. Unlike many existing methods, our method flexibly models classes using a variable number of components, which allows it to learn complex class boundaries. Also, unlike most of the methods, ours estimates the number of latent classes present in the data. Experiments on synthetic data and data sets from the UC Irvine machine learning repository show that, overall, our method achieves significant improvements in classification performance compared with the existing methods.",
author = "Jayaram Raghuram and Miller, {David Jonathan} and George Kesidis",
year = "2014",
month = "1",
day = "1",
doi = "10.1109/TNNLS.2013.2294459",
language = "English (US)",
volume = "25",
pages = "1520--1537",
journal = "IEEE Transactions on Neural Networks and Learning Systems",
issn = "2162-237X",
publisher = "IEEE Computational Intelligence Society",
number = "8",

}

Instance-level constraint-based semisupervised learning with imposed space-partitioning. / Raghuram, Jayaram; Miller, David Jonathan; Kesidis, George.

In: IEEE Transactions on Neural Networks and Learning Systems, Vol. 25, No. 8, 6695785, 01.01.2014, p. 1520-1537.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Instance-level constraint-based semisupervised learning with imposed space-partitioning

AU - Raghuram, Jayaram

AU - Miller, David Jonathan

AU - Kesidis, George

PY - 2014/1/1

Y1 - 2014/1/1

N2 - A new method for semisupervised learning from pairwise sample (must- and cannot-link) constraints is introduced. It addresses an important limitation of many existing methods, whose solutions do not achieve effective propagation of the constraint information to unconstrained samples. We overcome this limitation by constraining the solution to comport with a smooth (soft) class partition of the feature space, which necessarily entails constraint propagation and generalization to unconstrained samples. This is achieved via a parameterized mean-field approximation to the posterior distribution over component assignments, with the parameterization chosen to match the representation power of the chosen (generative) mixture density family. Unlike many existing methods, our method flexibly models classes using a variable number of components, which allows it to learn complex class boundaries. Also, unlike most of the methods, ours estimates the number of latent classes present in the data. Experiments on synthetic data and data sets from the UC Irvine machine learning repository show that, overall, our method achieves significant improvements in classification performance compared with the existing methods.

AB - A new method for semisupervised learning from pairwise sample (must- and cannot-link) constraints is introduced. It addresses an important limitation of many existing methods, whose solutions do not achieve effective propagation of the constraint information to unconstrained samples. We overcome this limitation by constraining the solution to comport with a smooth (soft) class partition of the feature space, which necessarily entails constraint propagation and generalization to unconstrained samples. This is achieved via a parameterized mean-field approximation to the posterior distribution over component assignments, with the parameterization chosen to match the representation power of the chosen (generative) mixture density family. Unlike many existing methods, our method flexibly models classes using a variable number of components, which allows it to learn complex class boundaries. Also, unlike most of the methods, ours estimates the number of latent classes present in the data. Experiments on synthetic data and data sets from the UC Irvine machine learning repository show that, overall, our method achieves significant improvements in classification performance compared with the existing methods.

UR - http://www.scopus.com/inward/record.url?scp=84904730137&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84904730137&partnerID=8YFLogxK

U2 - 10.1109/TNNLS.2013.2294459

DO - 10.1109/TNNLS.2013.2294459

M3 - Article

AN - SCOPUS:84904730137

VL - 25

SP - 1520

EP - 1537

JO - IEEE Transactions on Neural Networks and Learning Systems

JF - IEEE Transactions on Neural Networks and Learning Systems

SN - 2162-237X

IS - 8

M1 - 6695785

ER -