2002 · 167 citations · 13 references
Artificial IntelligenceEngineeringMachine LearningSemi-supervised Learning MethodBase ClassifierEnsemble MethodsData ScienceData MiningPattern RecognitionManagementStatisticsMultiple Classifier SystemPredictive AnalyticsKnowledge DiscoveryIntelligent ClassificationComputer ScienceData ClassificationClassification EnsemblesClassificationClassifier SystemCost-sensitive LearningCost-sensitive Machine LearningEnsemble Algorithm
The paper proposes ASSEMBLE, an adaptive semi‑supervised ensemble method that builds classification ensembles using both labeled and unlabeled data. ASSEMBLE iteratively assigns pseudo‑classes to unlabeled data with the current ensemble, then trains a new base classifier on both labeled and pseudo‑labeled data, effectively maximizing the classification margin without requiring a semi‑supervised base learner and compatible with any cost‑sensitive algorithm for binary or multiclass tasks. ASSEMBLE achieved top performance, winning the NIPS 2001 Unlabeled Data Competition with decision trees and demonstrating strong results on multiple benchmark datasets with decision trees and neural networks.
An adaptive semi-supervised ensemble method, ASSEMBLE, is proposed that constructs classification ensembles based on both labeled and unlabeled data. ASSEMBLE alternates between assigning "pseudo-classes" to the unlabeled data using the existing ensemble and constructing the next base classifier using both the labeled and pseudolabeled data. Mathematically, this intuitive algorithm corresponds to maximizing the classification margin in hypothesis space as measured on both the labeled and unlabeled of data. Unlike alternative approaches, ASSEMBLE does not require a semi-supervised learning method for the base classifier. ASSEMBLE can be used in conjunction with any cost-sensitive classification algorithm for both two-class and multi-class problems. ASSEMBLE using decision trees won the NIPS 2001 Unlabeled Data Competition. In addition, strong results on several benchmark datasets using both decision trees and neural networks support the proposed method.
13
Yuhai Wu, Vladimir Vapnik · Technometrics · 1999 · 26.9K citations
UCI Repository of machine learning databases
Catherine Blake · Medical Entomology and Zoology · 1998 · 10.5K citations
Experiments with a new boosting algorithm
Yoav Freund, Robert E. Schapire · 1996 · 7.6K citations
Combining labeled and unlabeled data with co-training
Avrim Blum, Tom M. Mitchell · 1998 · 5.6K citations · Full text