Research And Implementation On Multi-View Partial Multi-Label Learning Algorithm

In traditional multi-label learning framework,a common assumption is that each example in the data set used for training is accurately labeled with all relevant labels.However,due to the rapid growth of data in real world application,accurate labeling is difficult to be obtained and the cost is high,resulting in most of the final label information contains noisy labels.Therefore,partial multi-label learning is of great significant interest.In the framework of partial multi-label learning,the candidate label set of each instance contains at least one but unknown number of relevant labels and some noisy labels.The task of partial multi-label learning is to train a predication model using the candidate label sets containing redundant labels,so as to get the real labels.The main difficulty of partial multi-label learning is how to overcome the bias of model training caused by noisy labels in training candidate label sets.Recently,multi-view learning has been developed to deal with partial multi-label learning tasks.Although few multi-view partial multi-label learning methods have been proposed,all of them are designed under the full-view assumption.However,due to the difficulties in multi-view data collection,some views may not contain complete information in real task.The appearance of missing views will affect the performance of traditional partial multi-label learning algorithms.To solve this problem,we propose a novel incomplete multi-view partial multi-label learning framework,and two partial multi-label learning algorithms.In order to make better use of the information of multi-view data with missing views,this paper proposes Incomplete Multi-View Partial Multi-Label Learning algorithm（IMVPML）.In the training process,it makes use of incomplete multi-view feature representation and utilizes the low-rank and sparse decomposition scheme to remove the noisy labels.Specifically,IMVPML first learns a shared subspace across heterogenous incomplete views.Secondly,the ground-truth labels are obtained by the low-rank and sparse decomposition scheme.Thirdly,a graph Laplacian regularization is introduced to constrain the ground-truth labels and impose orthogonality constraints on the correlations between subspace.Finally,a predictive model is learned by shared subspace and disambiguation labels.Enormous experimental results demonstrate that the proposed method can achieve competitive performance in solving the problem of incomplete multi-view partial multi-label learning.This paper proposes Incomplete Multi-View Partial-Multi Labeling Learning via Feature Completion（IMVPML-FC）,in order to reconstruct full representation for incomplete view information.In the training process,the algorithm designs a model to recover the missing views,learns the similarity graph of all views,and utilize the low-rank and sparse decomposition scheme to remove noisy labels.In particular,IMVPML-FC firstly designs a new address representation model to recover the missing views,so as to obtain complete multi view data.Then,the similarity graph of features on all views is obtained by self-representation method.Finally,7)₁-norm regularization constrains recovering view noise and label noise.Many experiments are designed on a variety of multi-view data sets.The experimental results show that the proposed method has a good performance in multi view partial multi tag learning under missing views.