Learning Disentangled Representations for Generalized Multi-view Clustering

2026-05-15

IEEE Transactions on Pattern Analysis and Machine Intelligence

AI Summary

This work addresses the degradation of shared latent space quality and limited clustering performance in existing deep multi-view clustering methods, which stems from entangled view distributions during cross-view fusion. To overcome this, the paper proposes the Generalized Multi-view Autoencoder (GMAE), the first approach to introduce disentangled representation learning into multi-view clustering. GMAE employs a dual-path autoencoder to separate view-specific and view-common features, enhances feature discriminability through a cross-view adversarial discriminator, and leverages mutual information modulation to align distributions and prevent representation collapse. This design effectively preserves view complementarity and significantly improves cluster structure clarity. Extensive experiments demonstrate that GMAE consistently outperforms state-of-the-art methods across 13 benchmark datasets in both complete and incomplete multi-view clustering scenarios.

Abstract

Multi-View Clustering (MVC) has gained significant attention for its ability to leverage complementary information across diverse views. However, existing deep MVC methods often struggle with view-distribution entanglement during cross-view fusion, which hampers the quality of the shared latent space and leads to suboptimal Figures. To address this issue, we propose the Generalized Multi-view Auto-Encoder (GMAE), a framework designed to preserve cross-view complementarity through disentangled representation learning. Specifically, GMAE employs dual-path autoencoders to decouple source features into view-specific and view-common embeddings, facilitating the discovery of clearer clustering structures. We further construct cross-view adversarial discriminators to guide view-specific encoders in capturing more discriminative features. By strategically modulating mutual information, GMAE effectively aligns distributions and prevents representation collapse, ensuring the generation of robust, non-trivial embeddings. Comprehensive experiments on 13 benchmark datasets demonstrate that GMAE consistently outperforms state-of-the-art methods in both complete and incomplete MVC tasks. Our code implementation is available at the repository: https://github.com/obananas/GMAE.

Problem

• Multi-View Clustering
• View-Distribution Entanglement
• Disentangled Representations
• Cross-View Fusion
• Latent Space

Innovation

• disentangled representation
• multi-view clustering
• adversarial discriminator
• mutual information modulation
• view-specific embedding