.png){kind=icon}
Return
Multi-view Partial Multi-label Learning via Label Distribution
.png){kind=icon}
PRE
AI
DOI：.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Multi-view Partial Multi-Label Learning (MVPML) is a method that integrates data from multiple views to address multi-label classification tasks with noisy labels. Conventional MVPML methods typically adopt binary logical labels to represent sample-label relationships, thereby overlooking the varying degrees to which labels describe samples. Moreover, most existing methods assume that all features information of sample is valid data, but in practice, increasing feature dimensionality inevitably leads to feature redundancy. In order to solve the above problems, this paper proposes a multi-view partial multi-label learning algorithm based on label distribution, named MVPML_LD(Multi-View Partial Multi-Label Learning via Label Distribution), Specifically, First, MVPML_LD learns sample correlations and incorporates them into the classifier within each view, in order to mitigate the adverse effects of redundant features. Second, A pseudo-label distribution matrix is constructed by leveraging information from both the feature space and the original label space. Then, discrepancy between the pseudo-label and the true label distribution is then reduced, while the true label distribution is constrained using label correlations and sample correlations. Finally, the adaptive view-weighting mechanism is utilized to dynamically integrate the information from multiple views that has been filtered via sample correlations, and jointly optimize it with the label distribution within a unified predictive framework. Extensive experiments on multi-view partial multi-label datasets verify the effectiveness of the proposed method.
Keywords:
multi-view learning
partial multi-label learning
label noise
label distribution
redundant features
Journal
E
IF:
0.6
Papers:
205
Citations:
0
Organization
Cited Papers
No cited papers available