.png){kind=icon}
.png){kind=icon}
Return
Learning consistent representation for incomplete multi-view weak multi-label classification
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2025.131238.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In recent years, multi-view multi-label learning has become a popular interdisciplinary research topic. Building on this, this paper further explores the complex real-world problem of incomplete multi-view weak multi-label learning. In this problem, how to effectively utilize interrelationships across views to embed consistent learning objectives into models has been a key challenge. To address this, we propose a Learning Consistent Representation Network (LCRN) based on the Information Noise-Contrastive Estimation (InfoNCE) loss. This method transforms consistency learning between multiple views into a pseudo dictionary look-up problem. By matching the query with a positive key, it treats different views from the same sample as positive pairs, thereby promoting consistency learning across multiple views. Benefiting from the flexibility of splitting a single view into query and key representations, we can achieve mutual alignment in consistency learning between two views. Finally, we use missing view and missing label indicators to mitigate the negative performance impact caused by missing multi-views and weak multi-labels. After extensive experiments on five incomplete multi-view weak multi-label datasets, our model demonstrated competitive results.
Keywords:
multi-view learning
multi-label learning
consistent representation
InfoNCE loss
missing data handling
Journal
IF:
6.5
Papers:
2.5W
Citations:
6.5W