Robust Multi-label Learning

Multi-label learning refers to a sample may correspond to multiple classes and can be used in many practical application,such as text classification,gene function classification,scene classification,etc.In recent years,multi-label learning has received more and more attention and research due to its complexity and extensive application background.Traditional multi-label algorithms are often based on the assumption that the training data is noiseless or of a single type of noise,which leads to poor performance of the model on real complex noise data.In the real world,training data often contain different type of noise,such as attribute noise,label noise,or hybrid noise.Therefore,how to build a robust multi-label learning model is also a hot research direction in recent years.Based on the robust multi-label learning model,this paper proposes three robust multi-label learning model by introducing the framework of ensemble learning,hybrid noise model and multi-view learning respectively.The main contributions of this paper has the following three aspects:(1).Independence Regularized Multi-Label Ensemble,we propose an ensemble method to learn the basic classifier by introduce the Hilbert-Schmidt Independence Criterion(HSIC)to model the general independence of the different classifiers.And considering the different qualities of these classifiers,a weight a vector is learned to balance theses classifiers.(2).Hybrid Noise Oriented Multi-Label Learnin,we proposed the method which can simultaneously addresses feature and label noise by bi-sparsity regularization bridged with label enrichment.Specifically,the label enrichment matrix explores the underlying correlation among different classes which improve the noise labeling.Bridged with enriching label matrix,the structured sparsity is imposed to jointly handle the corrupted features and noisy labeling.(3).Latent Semantic Aware Multi-View Multi-Label classification,we propose a novel approach for multi-view multi-label learning based on matrix factorization to exploit complementarity among different views.Specifically,under the assumption that there exists a common representation across different vies,the uncovered latent patterns are enforced to be aligned across different views in kernel spaces.In this way,the latent semantic patterns underlying in data could be well uncovered and this enhances the reasonability of the common representation of multiple views.As a result,the consensus multi-view representation is obtained which encodes the complementarity and consistence of different views in latent semantic space.