Research And Implementation Of Weakly-Supervised Multi-Label Learning Algorithms Based On Low-Rank Matrix Constraints

Multi-label learning is a kind of framework in traditional machine learning tasks.In multi-label learning tasks,each instance of training data is associated with multiple labels simultaneously.However,in realistic scenarios,accurate label information is too difficult to obtain,so supervision information of the observed instances usually exists noise such as missing labels,noisy labels,and inaccuracy labels.Therefore,weakly-supervised multi-label learning has received widespread attention in recent years.Existing weakly-supervised multi-label learning algorithms mainly focus on three aspects: the first is multi-label learning problem with missing labels,and instances' label information in such problem is missing without redundant labels;the second is the problem with noisy labels,and the relevant algorithms concentrate on such problem with complete but redundant label information;the third is the problem with missing features,and the related methods mainly study the multi-label learning problem with missing features,including algorithms that focus on such problems with missing features and missing labels simultaneously.Nevertheless,in practical application,people inevitably have missing labels,wrong labels,and repeated labels when labeling samples.Thus it is more likely that missing and redundant labels exist in the label information of research instances simultaneously,but few methods pay attention to such issue.Meanwhile,occlusion,illumination and low resolution usually lead to the acquired instance features being noisy rather than just feature absence,which may reduce the robustness of existing learning models.Aiming at two different problems mentioned above,this paper proposes two corresponding weakly-supervised multi-label learning algorithms.A cost-sensitive label ranking approach with low-rank and sparse constraints for weakly-supervised multi-label learning is proposed.This approach deals with the problem of missing labels and noisy labels coexisting in the observed label information to enrich the missing labels and remove the noisy labels simultaneously.Unlike most existing studies that an indicator matrix needs to be given in advance which may not be available in reality,a label confidence matrix is constructed to reflect the relevance between the labels and the corresponding instances,and then the relevance ordering of all possible labels including both missing and noisy labels on each instance is optimized by minimizing a cost-sensitive ranking loss.By considering the dependencies in both feature space and label space,we exploit the dual low-rank regularization terms to capture the corresponding correlations.Afterwards,noticing the fact that both missing and noisy labels are rare,the sparse regularization term is encoded to constrain such noisy information to be sparse.Comprehensive experimental results demonstrate the effectiveness of the proposed method.An inductive framework based on matrix low-rank and sparse constraints for weakly-supervised multi-label learning is presented.This method addresses the multi-label learning problem with both noisy features and missing labels.Specifically,we first decompose the observed feature matrix into an ideal feature matrix and an outlier matrix.Considering that similar instances usually share similar visual characteristics,we constrain the ideal feature matrix to be low-rank.Meanwhile,a reasonable assumption is that the noise is sparse compared with the feature matrix,which leads outlier matrix to be sparse.In addition,a linear self-recovery model is adopted to reconstruct the incomplete label assignment matrix by exploiting label correlations.Finally,the desired model is trained on the ideal feature matrix and the refined label matrix.Extensive experimental results demonstrate that the proposed model can achieve superior and comparable performance against state-of-the-art methods.