.png){kind=icon}
Return
Missing Multi-Label Learning With TSK Fuzzy System and Adaptive Graph
.png){kind=icon}
PRE
AI
DOI：10.1109/tfuzz.2026.3716853.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Multi-label learning with missing labels is challenging in real-world applications and has received considerable attention in recent years. Most existing methods employ label correlations to address the problem of missing labels. However, these methods directly learn label correlations from the original missing labels matrix, which increases erroneous information. Although some approaches use data augmentation to improve label recovery, they rarely explore the relationship between recovered labels and the feature space.Takagi-Sugeno-Kang (TSK) fuzzy system has strong nonlinear modeling capabilities and interpretability, making it a promising model for addressing the missing labels problem. Therefore, we present a missing multi-label learning method with TSK fuzzy system and adaptive graph. Specifically, the TSK fuzzy system transforms the original feature space into a fuzzy space and enhances feature representations for instances with missing labels. Subsequently, the adaptive graph is used to obtain the label correlations and then applied to recover missing labels, thereby learning the nonlinear relationship between the recovered labels and the fuzzy representations. Moreover, the label manifold structure is employed to constrain the predicted labels to have the same label correlation as the recovered label matrix, guiding the prediction of unknown instances for generalization improvement. Experimental results demonstrate that our proposed approach outperforms state-of-the-art methods across datasets from different domains.
Keywords:
Adaptive graph
missing labels
multi-label learning
TSK fuzzy system
Journal
IF:
11.9
Papers:
5.0K
Citations:
2.9W
Organization
Cited Papers
No cited papers available