Neural Information Processing Systems · 2012 · 220 citations · 21 references
ComputingComputer ScienceData MiningFeature EngineeringMachine LearningDimension ReductionDimensionality ReductionFeature SelectionImage ClassificationKernel MethodsRepresentation LearningBenchmark DatasetsCompressive SensingFeature Dimensionality ReductionHamming DistanceMinimizationSingular Value DecompositionSparse Feature RepresentationSubspace Learning
Label space dimension reduction (LSDR) is an efficient and effective paradigm for multi-label classification with many classes. Existing approaches to LSDR, such as compressive sensing and principal label space transformation, exploit only the label part of the dataset, but not the feature part. In this paper, we propose a novel approach to LSDR that considers both the label and the feature parts. The approach, called conditional principal label space transformation, is based on minimizing an upper bound of the popular Hamming loss. The minimization step of the approach can be carried out efficiently by a simple use of singular value decomposition. In addition, the approach can be extended to a kernelized version that allows the use of sophisticated feature combinations to assist LSDR. The experimental results verify that the proposed approach is more effective than existing ones to LSDR across many real-world datasets.
21
Mark Hall, Eibe Frank, Geoffrey Holmes et al. · ACM SIGKDD Explorations Newsletter · 2009 · 17.8K citations
Colin Goodall, Ian T. Jolliffe · Technometrics · 1988 · 8.7K citations
RELATIONS BETWEEN TWO SETS OF VARIATES
Harold Hotelling · Biometrika · 1936 · 5.4K citations
The Approximation of One Matrix by Another of Lower Rank
Carl Eckart, Gale Young · Psychometrika · 1936 · 3.8K citations
Learning multi-label scene classification
Matthew Boutell, Jiebo Luo, Xipeng Shen et al. · Pattern Recognition · 2004 · 2.3K citations