Multi-domain Feature Integration based Trusted Partial Multi-view Incomplete Multi-label Learning.

Area of Science:

• Computer Science
• Machine Learning
• Artificial Intelligence

Background:

• Partial Multi-view Incomplete Multi-label Learning (PMvIMlL) addresses real-world data challenges with missing views and labels.
• Existing PMvIMlL methods struggle with quantifying predictive uncertainty, limited feature representations, and suboptimal fusion quality due to cross-view divergence.

Purpose of the Study:

• To propose MFIT, a framework that overcomes limitations in existing PMvIMlL methods.
• To enhance predictive accuracy and decision reliability in incomplete multi-label learning tasks.

Main Methods:

• MFIT utilizes a Multi-domain Feature Integration approach.
• Key components include Feature Enhancement via Frequency-Domain Integration (FE-FDI) and Evidential Fusion with Cross-view Decision Consistency (EF-CDC).
• FE-FDI incorporates frequency-domain information alongside spatial features; EF-CDC improves fusion by managing cross-view conflicts.

Main Results:

• MFIT demonstrated substantial and consistent performance improvements across six benchmark datasets.
• The framework effectively enhanced both prediction accuracy and decision reliability.
• FE-FDI successfully enriched data representations by integrating global frequency patterns.
• EF-CDC mitigated issues related to multi-view decision conflicts during fusion.

Conclusions:

• MFIT offers a robust solution for Partial Multi-view Incomplete Multi-label Learning.
• The proposed FE-FDI and EF-CDC modules significantly advance the state-of-the-art in handling dual missing data.
• MFIT provides more reliable and accurate predictions in complex multi-label learning scenarios.