Reliable Representation Learning for Incomplete Multi-View Missing Multi-Label Classification.

Area of Science:

• Computer Science
• Machine Learning
• Artificial Intelligence

Background:

• Multi-view multi-label classification is an emerging area combining multi-view learning and multi-label classification.
• Existing multi-view contrastive learning methods often incorrectly separate similar samples, and many multi-view multi-label methods fail with incomplete data.
• There is a need for robust methods that handle missing views and labels while preserving data structure.

Purpose of the Study:

• To propose a novel network, RANK, for incomplete multi-view missing multi-label classification.
• To address the limitations of existing methods in handling negative pair separation and data incompleteness.
• To improve the accuracy and robustness of multi-view multi-label classification.

Main Methods:

• Developed a label-driven multi-view contrastive learning strategy to maintain intra-view structure and align cross-view information.
• Introduced a quality-aware subnetwork for dynamic view quality scoring, overcoming fixed view-level weights.
• Utilized label correlation within a multi-label cross-entropy loss for enhanced discriminative power.

Main Results:

• The proposed RANK network effectively handles both complete and incomplete multi-view multi-label datasets.
• RANK demonstrates superior performance compared to existing state-of-the-art methods in extensive experiments.
• The label-driven contrastive learning and quality-aware subnetwork contribute to improved classification accuracy.

Conclusions:

• RANK offers a significant advancement in multi-view multi-label classification, particularly for incomplete datasets.
• The method's ability to handle missing views and labels makes it broadly applicable.
• RANK's innovative approach enhances feature representation and classification performance.