Incomplete Multi-View Multi-Label Classification via Shared Codebook and Fused-Teacher Self-Distillation

Forum topic · 小凯 · 2026-04-07

Summary

This paper addresses the dual-missing scenario in multi-view multi-label classification, where both views and labels are incomplete. Existing approaches rely on contrastive learning or information bottleneck theory, but loss-based alignment without explicit structural constraints limits the capture of stable, discriminative shared semantics. The authors propose learning discrete consistent representations through a multi-view shared codebook with cross-view reconstruction, which aligns different views within limited shared codebook embeddings and reduces feature redundancy. At the decision level, a weight estimation method evaluates each view's ability to preserve label correlation structures to improve fused prediction quality. A fused-teacher self-distillation framework further lets the fused prediction guide training of view-specific classifiers, feeding global knowledge back into single-view branches to enhance generalization under missing labels. Experiments on five benchmark datasets demonstrate effectiveness; code is available at https://github.com/xuy11/SCSD.

Paper Overview

Research Area: Computer Vision (CV) Authors: Xu Yan, Jun Yin, Shiliang Sun

Abstract

Although multi-view multi-label learning has been extensively studied, research on the dual-missing scenario, where both views and labels are incomplete, remains largely unexplored. Existing methods mainly rely on contrastive learning or information bottleneck theory to learn consistent representations under missing-view conditions, but loss-based alignment without explicit structural constraints limits the ability to capture stable and discriminative shared semantics. To address this issue, we introduce a more structured mechanism for consistent representation learning: we learn discrete consistent representations through a multi-view shared codebook and cross-view reconstruction, which naturally align different views within the limited shared codebook embeddings and reduce feature redundancy. At the decision level, we design a weight estimation method that evaluates the ability of each view to preserve label correlation structures, assigning weights accordingly to enhance the quality of the fused prediction. In addition, we introduce a fused-teacher self-distillation framework, where the fused prediction guides the training of view-specific classifiers and feeds the global knowledge back into the single-view branches, thereby enhancing the generalization ability of the model under missing-label conditions. The effectiveness of our proposed method is thoroughly demonstrated through extensive comparative experiments with advanced methods on five benchmark datasets. Code is available at https://github.com/xuy11/SCSD.

Key Contributions

Shared codebook representation learning: Discrete consistent representations learned via a multi-view shared codebook and cross-view reconstruction, aligning views within limited codebook embeddings while reducing feature redundancy.
View weight estimation: A decision-level method that measures each view's ability to preserve label correlation structures and assigns weights to improve fused prediction quality.
Fused-teacher self-distillation: The fused prediction acts as a teacher guiding view-specific classifiers, feeding global knowledge back into single-view branches to improve generalization with missing labels.

Results

Extensive comparative experiments against state-of-the-art methods on five benchmark datasets validate the proposed approach.

Code: https://github.com/xuy11/SCSD