Semisupervised learning of classifiers: theory, algorithms, and their application to human-computer interaction
IEEE Transactions on Pattern Analysis and Machine IntelligencePublished 19 October 2004
Ira L. Cohen, Fábio Gagliardi Cozman, Nicu Sebe, Marcelo César Cirelo, Thomas S. Huang
Citations252
SJR quartileQ1
SJR score3.91
SNIP5.99
Generate an AI Snapshot to get a quick, structured summary of this paper.
Study Snapshot
ObjectiveStudy objective
MethodsResearch methodology
PopulationPopulation studied
Sample sizeSample sizes
OutcomesStudy outcomes here
ResultsStudy results comes here
LimitationsResearch study limitations comes here
A concise AI-generated summary of the paper will appear here once you click Generate AI Snapshot.
TL;DR
A new analysis is provided that shows under what conditions unlabeled data can be used in learning to improve classification performance, and how the resulting algorithms are successfully employed in two applications related to human-computer interaction and pattern recognition: facial expression recognition and face detection.
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
Computer Science
Journal of the Royal Statistical Society Series B (Statistical Methodology)Maximum Likelihood from Incomplete Data Via the <i>EM</i> Algorithm
49,657 Citations1977A. P. Dempster, N. M. Laird +1 more
The Journal of Chemical PhysicsEquation of State Calculations by Fast Computing Machines
37,074 Citations1953N. Metropolis, Arianna W. Rosenbluth +3 more
Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference
16,927 Citations1988Judea Pearl
The author provides a coherent explication of probability as a language for reasoning with partial belief and offers a unifying perspective on other AI approaches to uncertainty, such as the Dempster-Shafer formalism, truth maintenance systems, and nonmonotonic logic.
Combining labeled and unlabeled data with co-training
5,604 Citations1998Avrim Blum, Tom M. Mitchell
Lecture notes in statisticsCausation, Prediction, and Search
4,740 Citations1993Peter Spirtes, Clark Glymour +1 more
The authors axiomatize the connection between causal structure and probabilistic independence, explore several varieties of causal indistinguishability, formulate a theory of manipulation, and develop asymptotically reliable procedures for searching over equivalence classes of causal models.
Machine LearningBayesian Network Classifiers
4,738 Citations1997Nir Friedman, Dan Geiger +1 more
Tree Augmented Naive Bayes (TAN) is single out, which outperforms naive Bayes, yet at the same time maintains the computational simplicity and robustness that characterize naive Baye.
The MIT Press eBooksCausation, Prediction, and Search
4,470 Citations2001Peter Spirtes, Clark Glymour +1 more
IEEE Transactions on Pattern Analysis and Machine IntelligenceDetecting faces in images: a survey
3,429 Citations2002Shuicheng Yan, David Kriegman +1 more
Stochastic modelling and applied probabilityA Probabilistic Theory of Pattern Recognition
3,277 Citations1996Luc Devroye, László Györfi +1 more
The Bayes Error and Vapnik-Chervonenkis theory are applied as guide for empirical classifier selection on the basis of explicit specification and explicit enforcement of the maximum likelihood principle.
Machine LearningText Classification from Labeled and Unlabeled Documents using EM
2,749 Citations2000Kamal Nigam, Andrew Kachites McCallum +2 more
This paper shows that the accuracy of learned text classifiers can be improved by augmenting a small number of labeled training documents with a large pool of unlabeled documents, and presents two extensions to the algorithm that improve classification accuracy under these conditions.
Journal of the American Statistical AssociationDiscriminant Analysis and Statistical Pattern Recognition.
2,671 Citations1993Douglas M. Hawkins, Geoffrey J. McLachlan
Comprehensive database for facial expression analysis
2,582 Citations2002Takeo Kanade, Jeffrey F. Cohn +1 more
The problem space for facial expression analysis is described, which includes level of description, transitions among expressions, eliciting conditions, reliability and validity of training and test data, individual differences in subjects, head orientation and scene complexity image characteristics, and relation to non-verbal behavior.
IEEE Transactions on Pattern Analysis and Machine IntelligenceAutomatic analysis of facial expressions: the state of the art
1,764 Citations2000M. Pantie, Léon Rothkrantz
The capability of the human visual system with respect to these problems is discussed, and it is meant to serve as an ultimate goal and a guide for determining recommendations for development of an automatic facial expression analyzer.
Scaling up the accuracy of Naive-Bayes classifiers: a decision-tree hybrid
1,363 Citations1996Ron Kohavi
A new algorithm, NBTree, is proposed, which induces a hybrid of decision-tree classifiers and Naive-Bayes classifiers: the decision-Tree nodes contain univariate splits as regular decision-trees, but the leaves contain Naïve-Bayesian classifiers.
Mathematics of Operations ResearchCooling Schedules for Optimal Annealing
1,232 Citations1988Bruce Hajek
A Monte Carlo optimization technique called “simulated annealing” is a descent algorithm modified by random ascent moves in order to escape local minima which are not global minima.
International Statistical ReviewBayesian Graphical Models for Discrete Data
1,115 Citations1995David Madigan, Jeremy York +1 more
Psychological BulletinStrong evidence for universals in facial expressions: A reply to Russell's mistaken critique.
1,053 Citations1994Paul Ekman
The evidence from both literate and preliterate cultures is overwhelming in support of universals in facial expressions, and A.J. Russell's qualms about other aspects of the design of the studies of literate cultures have no merit.
Data Mining and Knowledge DiscoveryOn Bias, Variance, 0/1—Loss, and the Curse-of-Dimensionality
1,029 Citations1997Jerome H. Friedman
This work candramatically mitigate the effect of the bias associated with some simpleestimators like “naive” Bayes, and the bias induced by the curse-of-dimensionality on nearest-neighbor procedures.
Artificial IntelligenceLearning Bayesian networks from data: An information-theory based approach
803 Citations2002Jie Cheng, Russell Greiner +3 more
Algorithms that use an information-theoretic analysis to learn Bayesian network structures from data, requiring only polynomial numbers of conditional independence tests in typical cases are provided.
Semi-Supervised Support Vector Machines
793 Citations1998Kristin P. Bennett, Ayhan Demiriz
A general S3VM model is proposed that minimizes both the misclassification error and the function capacity based on all the available data that can be converted to a mixed-integer program and then solved exactly using integer programming.
IEEE Transactions on Geoscience and Remote SensingThe effect of unlabeled samples in reducing the small sample size problem and mitigating the Hughes phenomenon
572 Citations1994B.M. Shahshahani, D. A. Landgrebe
By using additional unlabeled samples that are available at no extra cost, the performance may be improved, and therefore the Hughes phenomenon can be mitigated and therefore more representative estimates can be obtained.
Enhancing Supervised Learning with Unlabeled Data
459 Citations2000Sally A. Goldman, Yan Zhou
A new semi-supervised learning method called co-learning that is designed to use unlabeled data to enhance standard supervised learning algorithms to leverage off the fact that they have different representations of the hypotheses and are likely to detect different patterns in labeled data.
Comparing Bayesian network classifiers
341 Citations1999Jie Cheng, Russell Greiner
Layered representations for human activity recognition
306 Citations2003Nuria Oliver, Eric Horvitz +1 more
The use of representation in a system that diagnoses states of a user's activity based on real-time streams of evidence from video, acoustic, and computer interactions is described.
Neural Information Processing SystemsA Mixture of Experts Classifier with Learning Based on Both Labelled and Unlabelled Data
289 Citations1996David J. Miller, Hasan S. Uyar
A classifier structure and learning algorithm that make effective use of unlabelled data to improve performance and is a "mixture of experts" structure that is equivalent to the radial basis function (RBF) classifier, but unlike RBFs, is amenable to likelihood-based training.
Semi-supervised learning of mixture models
192 Citations2003Fábio Gagliardi Cozman, Ira L. Cohen +1 more
This paper analyzes the performance of semi-supervised learning of mixture models and shows that unlabeled data can lead to an increase in classification error even in situations where additional labeled data would decrease classification error.
The Role of Unlabeled Data in Supervised Learning
150 Citations2004Tom M. Mitchell
It is argued that models of human and animal learning should consider more strongly the potential role of unlabeled data, and that many natural learning problems fit the problem class identified in this paper.
Journal of the American Statistical AssociationNormal Discrimination with Unclassified Observations
144 Citations1978Terence J. O’Neill
Machine LearningStructural Extension to Logistic Regression: Discriminative Parameter Learning of Belief Net Classifiers
136 Citations2005Russell Greiner, Xiaoyuan Su +2 more
Empirical evidence is presented that ELR produces effective classifiers, often superior to the ones produced by the standard “generative” algorithms, especially in common situations where the given BN-structure is incorrect.
Unlabeled Data Can Degrade Classification Performance of Generative Classifiers
130 Citations2002Fábio Gagliardi Cozman, Ira L. Cohen
It is shown that unlabeled data can degrade the performance of a classifier when there are discrepancies between modeling assumptions used to build the classifier and the actual model that generates the data.
Joint processing of audio-visual information for the recognition of emotional expressions in human-computer interaction
108 Citations2000Thomas S. Huang, Lawrence Shao-Hsien Chen
This thesis addresses the problem of detecting human emotional expressions by computer from the voice and facial motions of the user and proposes a complementary relationship between audio and video.
Combining Labeled and Unlabeled Data for MultiClass Text Categorization
104 Citations2002Rayid Ghani
This paper develops a framework to incorporate unlabeled data in the Error-Correcting Output Coding (ECOC) setup by first decomposing multiclass problems into multiple binary problems and then using Co-Training to learn the individual binary classi cation problems.
Learning from a mixture of labeled and unlabeled examples with parametric side information
97 Citations1995Joel Ratsaby, Santosh S. Venkatesh
The tradeoff between labeled and unlabeled sample complexities in learning is investigated and pendent sampling from the distribution of pairs (x, y) is shown.
WORLD SCIENTIFIC eBooksPattern Recognition
96 Citations2001Sankar K. Pal, Amita Pal
Pattern recognition - evolution of methodologies and data mining, A.K. Pal and S.S. Pal adaptive stochastic algorithms for pattern classification, M.M. Mardia decision trees for classification - a review and some new results, R. Skowron and R.N. Zwir.
Lecture notes in computer sciencePositive and Unlabeled Examples Help Learning
94 Citations1999Francesco de Comité, François Denis +2 more
This work addresses the problem of learning with the help of positive and unlabeled data given a small number of labeled examples and presents both theoretical and empirical arguments showing that learning algorithms can be improved by the use of both unlabeling and positive data.
Facial expression recognition from video sequences
87 Citations2003Ira L. Cohen, Nicu Sebe +3 more
This work introduces a tree-augmented naive-Bayes (TAN) classifier that learns the dependencies between facial features and provides an algorithm for finding the best TAN structure.
National Conference on Artificial IntelligenceStructural extension to logistic regression: discriminative parameter learning of belief net classifiers
87 Citations2002Russell Greiner, Wei Zhou
Probabilistic Modeling for Face Orientation Discrimination: Learning from Labeled and Unlabeled Data
87 Citations1998Shumeet Baluja
BiometrikaThe efficiency of a linear discriminant function based on unclassified initial samples
77 Citations1978S. Gànesalingam, Geoffrey J. McLachlan
Connected vibrations: a modal analysis approach for non-rigid motion tracking
75 Citations2002Tao Hai, Thomas S. Huang
A new framework for tracking non-rigid motion in image sequences as the deformations of connected 2D membrane patches using stochastic filtering technique to improve the robustness of the tracker and promising results on both natural and synthetic facial motion sequences are demonstrated.
Proceedings of the IEEEBoosted learning in dynamic bayesian networks for multimodal speaker detection
51 Citations2003A. Garg, Vladimir Pavlović +1 more
A supervised learning framework for Bayesian networks, which is based on the Adaboost algorithm of Schapire and Freund, is developed and tested in the context of a novel multimodal HCI application: a speech-based command and control interface for a Smart Kiosk.
Learning in natural language
51 Citations2000Dan Roth
This article summarizes work on developing a learning theory account for the major learning and statistics based approaches used in natural language processing and shows that they can all be explained using a single distribution free inductive principle related to the pac model of learning.
Journal of the American Statistical AssociationLinear Discriminant Analysis with Misallocation in Training Samples
42 Citations1984Raj S. Chhikara, Jim McKeon
Model Selection Criteria for Learning Belief Nets: An Empirical Comparison
38 Citations2000Tim Van Allen, Russell Greiner
Continuation methods for mixing heterogeneous sources
31 Citations2002Adrian Corduneanu, Tommi Jaakkola
Pattern RecognitionLearning with imperfectly labeled patterns
28 Citations1980C. B. Chittineni
To gain an understanding of the learning process, the author derives expressions for success probability as a function of training time for a one-dimensional increment error correction classifier with imperfect labels.
IEEE Transactions on ComputersOn the Asymptotic Improvement in the Out- come of Supervised Learning Provided by Additional Nonsupervised Learning
26 Citations1970David B. Cooper, John H. Freeman
Simple mathematical expressions are derived for the improvement in supervised learning provided by additional nonsupervised learning when the number of learning samples is large so that asymptotic approximations are appropriate.
The relative value of labeled and unlabeled samples in pattern recognition
24 Citations2005V. Castelli, Thomas M. Cover
It is argued that labeled samples are exponentially more valuable than unlabeled samples and identify the (exponent as the Bhatthacharyya distance).
Layered Representations for Recognizing Office Activity
22 Citations2002Nuria Oliver, Eric Horvitz +1 more
The use of layered probabilistic representations using Hidden Markov Models for performing sensing, learning, and inference at multiple levels of temporal granularity in an office-awareness application is described.
Elsevier eBooksDiscriminant Analysis When Scale Contamination Is Present in the Initial Sample
16 Citations1977Susan W. Ahmed, Peter A. Lachenbruch
Pattern RecognitionEfficiency of discriminant analysis when initial samples are classified stochastically
14 Citations1990T. Krishnan, Subhas C. Nandy
The Efron efficiency of this procedure compared to the situation where the initial classification is done deterministically and correctly is studied, which concludes that stochastic supervision contains a great deal of information on the discriminant function.
Semisupervised learning of classifiers with application to human-computer interaction
13 Citations2003Thomas S. Huang, Ira L. Cohen
This work shows how the theory and algorithms applied in this work are successfully applied in two applications related to human-computer interaction: facial expression recognition and face detection.
Pattern RecognitionEfficiency of logistic-normal stochastic supervision
2 Citations1990T. Krishnan, Swagata Nandy
The Efron efficiency of this stochastic supervision scheme, in which the logarithm of odds ratio of the supervisor classification has a normal distribution and is independent of the feature vector, is computed.