Semisupervised learning of classifiers: Theory, algorithms, and their application to human-computer interaction

Ira Cohen; Fabio G. Cozman; Nicu Sebe; Marcelo C. Cirelo; Thomas S. Huang

doi:10.1109/TPAMI.2004.127

Semisupervised learning of classifiers: Theory, algorithms, and their application to human-computer interaction

Ira Cohen, Fabio G. Cozman, Nicu Sebe, Marcelo C. Cirelo, Thomas S. Huang

Coordinated Science Lab

Research output: Contribution to journal › Article › peer-review

Abstract

Automatic classification is one of the basic tasks required in any pattern recognition and human computer interaction application. In this paper, we discuss training probabilistic classifiers with labeled and unlabeled data. We provide a new analysis that shows under what conditions unlabeled data can be used in learning to improve classification performance. We also show that, if the conditions are violated, using unlabeled data can be detrimental to classification performance. We discuss the implications of this analysis to a specific type of probabilistic classifiers, Bayesian networks, and propose a new structure learning algorithm that can utilize unlabeled data to improve classification. Finally, we show how the resulting algorithms are successfully employed in two applications related to human-computer interaction and pattern recognition: facial expression recognition and face detection.

Original language	English (US)
Pages (from-to)	1553-1567
Number of pages	15
Journal	IEEE transactions on pattern analysis and machine intelligence
Volume	26
Issue number	12
DOIs	https://doi.org/10.1109/TPAMI.2004.127
State	Published - Dec 2004

Keywords

Bayesian network classifiers
Face detection
Facial expression recognition
Generative models
Semisupervised learning
Unlabeled data

ASJC Scopus subject areas

Software
Computer Vision and Pattern Recognition
Computational Theory and Mathematics
Artificial Intelligence
Applied Mathematics

Access to Document

10.1109/TPAMI.2004.127

Fingerprint Dive into the research topics of 'Semisupervised learning of classifiers: Theory, algorithms, and their application to human-computer interaction'. Together they form a unique fingerprint.

Cite this

Semisupervised learning of classifiers : Theory, algorithms, and their application to human-computer interaction. / Cohen, Ira; Cozman, Fabio G.; Sebe, Nicu; Cirelo, Marcelo C.; Huang, Thomas S.

In: IEEE transactions on pattern analysis and machine intelligence, Vol. 26, No. 12, 12.2004, p. 1553-1567.

Research output: Contribution to journal › Article › peer-review

@article{bf06ae92a16c45ffa59412143fe6315d,

title = "Semisupervised learning of classifiers: Theory, algorithms, and their application to human-computer interaction",

abstract = "Automatic classification is one of the basic tasks required in any pattern recognition and human computer interaction application. In this paper, we discuss training probabilistic classifiers with labeled and unlabeled data. We provide a new analysis that shows under what conditions unlabeled data can be used in learning to improve classification performance. We also show that, if the conditions are violated, using unlabeled data can be detrimental to classification performance. We discuss the implications of this analysis to a specific type of probabilistic classifiers, Bayesian networks, and propose a new structure learning algorithm that can utilize unlabeled data to improve classification. Finally, we show how the resulting algorithms are successfully employed in two applications related to human-computer interaction and pattern recognition: facial expression recognition and face detection.",

keywords = "Bayesian network classifiers, Face detection, Facial expression recognition, Generative models, Semisupervised learning, Unlabeled data",

author = "Ira Cohen and Cozman, {Fabio G.} and Nicu Sebe and Cirelo, {Marcelo C.} and Huang, {Thomas S.}",

note = "Funding Information: The authors would like to acknowledge Alex Bronstein and Marsha Duro of HP-Labs for their support and discussions on the analysis of semisupervised learning. Portions of this work have been done at HP-Labs, Palo-Alto. Ira Cohen has been funded by an HP Fellowship. Financial support has also been granted by HP-Brazil through the Eldorado Research Institute. The authors would like to thank Jilin Tu for his work on the face tracking, Roy Wang and Ming-Hsuan Yang for their help with the face detection database, and Jeffery Cohn and Larry Chen for the facial expression databases. They would like to thank Moises Goldszmidt, Ashutosh Garg, Ismenia Blavatsky, and Michael Lew for discussions on various parts of this work and the reviewers for their helpful comments. They coded our own Naive Bayes and TAN classifiers in the Java language, using the libraries of the JavaBayes system (freely available at http://www.cs.cmu.edu/~javabayes). This work has been partially funded by US National Science Foundation Grant IIS 00-85980.",

year = "2004",

month = dec,

doi = "10.1109/TPAMI.2004.127",

language = "English (US)",

volume = "26",

pages = "1553--1567",

journal = "IEEE Transactions on Pattern Analysis and Machine Intelligence",

issn = "0162-8828",

publisher = "IEEE Computer Society",

number = "12",

}

TY - JOUR

T1 - Semisupervised learning of classifiers

T2 - Theory, algorithms, and their application to human-computer interaction

AU - Cohen, Ira

AU - Cozman, Fabio G.

AU - Sebe, Nicu

AU - Cirelo, Marcelo C.

AU - Huang, Thomas S.

N1 - Funding Information: The authors would like to acknowledge Alex Bronstein and Marsha Duro of HP-Labs for their support and discussions on the analysis of semisupervised learning. Portions of this work have been done at HP-Labs, Palo-Alto. Ira Cohen has been funded by an HP Fellowship. Financial support has also been granted by HP-Brazil through the Eldorado Research Institute. The authors would like to thank Jilin Tu for his work on the face tracking, Roy Wang and Ming-Hsuan Yang for their help with the face detection database, and Jeffery Cohn and Larry Chen for the facial expression databases. They would like to thank Moises Goldszmidt, Ashutosh Garg, Ismenia Blavatsky, and Michael Lew for discussions on various parts of this work and the reviewers for their helpful comments. They coded our own Naive Bayes and TAN classifiers in the Java language, using the libraries of the JavaBayes system (freely available at http://www.cs.cmu.edu/~javabayes). This work has been partially funded by US National Science Foundation Grant IIS 00-85980.

PY - 2004/12

Y1 - 2004/12

N2 - Automatic classification is one of the basic tasks required in any pattern recognition and human computer interaction application. In this paper, we discuss training probabilistic classifiers with labeled and unlabeled data. We provide a new analysis that shows under what conditions unlabeled data can be used in learning to improve classification performance. We also show that, if the conditions are violated, using unlabeled data can be detrimental to classification performance. We discuss the implications of this analysis to a specific type of probabilistic classifiers, Bayesian networks, and propose a new structure learning algorithm that can utilize unlabeled data to improve classification. Finally, we show how the resulting algorithms are successfully employed in two applications related to human-computer interaction and pattern recognition: facial expression recognition and face detection.

AB - Automatic classification is one of the basic tasks required in any pattern recognition and human computer interaction application. In this paper, we discuss training probabilistic classifiers with labeled and unlabeled data. We provide a new analysis that shows under what conditions unlabeled data can be used in learning to improve classification performance. We also show that, if the conditions are violated, using unlabeled data can be detrimental to classification performance. We discuss the implications of this analysis to a specific type of probabilistic classifiers, Bayesian networks, and propose a new structure learning algorithm that can utilize unlabeled data to improve classification. Finally, we show how the resulting algorithms are successfully employed in two applications related to human-computer interaction and pattern recognition: facial expression recognition and face detection.

KW - Bayesian network classifiers

KW - Face detection

KW - Facial expression recognition

KW - Generative models

KW - Semisupervised learning

KW - Unlabeled data

UR - http://www.scopus.com/inward/record.url?scp=9244243116&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=9244243116&partnerID=8YFLogxK

U2 - 10.1109/TPAMI.2004.127

DO - 10.1109/TPAMI.2004.127

M3 - Article

C2 - 15573817

AN - SCOPUS:9244243116

VL - 26

SP - 1553

EP - 1567

JO - IEEE Transactions on Pattern Analysis and Machine Intelligence

JF - IEEE Transactions on Pattern Analysis and Machine Intelligence

SN - 0162-8828

IS - 12

ER -