Label-Guided Representation Learning for Incomplete Multi-View Multi-Label Classification

Incomplete multi-view multi-label classification addresses scenarios where views and labels are partially missing. While existing methods treat labels solely as supervision signals, they overlook the semantic structure inherent in partial annotations. We propose Label-Guided Representation Learning (LGRL) that systematically exploits label semantics as structural priors throughout learning. Our framework constructs a semantic-informed mixture prior via learnable category prototypes to guide representation extraction, and introduces category-specific conditional posteriors where prototypes act as Bayesian experts for multi-view fusion. We further derive a principled label-driven information bottleneck objective balancing reconstruction sufficiency with cross-view consistency, enabling category-conditional reasoning. Extensive experimental results demonstrate the effectiveness of LGRL across benchmark datasets as well as real-world applications in sports analytics and medical imaging.

Many real-world objects can be described from several angles at once and carry several labels at the same time. A photo, for instance, comes with its colors, textures, and shapes, and might be tagged "beach," "ocean," and "sky" together. In practice much of this information is missing. Some views are unavailable, and many tags are never filled in. Most existing methods treat the labels that do exist merely as answers to check against, ignoring the structure hidden in them, such as the fact that "beach" makes "ocean" likely but "mountain" unlikely. We built a method that puts these partial labels to work as guidance. For each category we learn a "prototype" that captures what that category typically looks like. These prototypes shape how the system represents data and how it combines the available views, adapting to which category is being judged, since different categories rely on different clues. Across six standard datasets, as well as basketball player analysis and chest X-ray screening, our approach consistently outperforms existing methods, and its advantage grows when data is severely incomplete.