Balanced Multi-view Clustering

TL;DR

This work tackles imbalanced information across views in unsupervised multi-view clustering by introducing Balanced Multi-view Clustering (BMvC) with View-specific Contrastive Regularization (VCR). BMvC preserves view-specific and joint clustering signals through a shared representation $\mathbf{F}$ while injecting a contrastive objective to balance gradient updates across views, supported by a theoretical analysis of gradient modulation. Empirical results on eight benchmark datasets show that BMvC consistently outperforms state-of-the-art MVC methods, confirming better exploitation of view-specific patterns and exploration of view-invariance cues. The approach offers a practical, robust solution for leveraging heterogeneous multi-view data in clustering tasks with varying view quality.

Abstract

Multi-view clustering (MvC) aims to integrate information from different views to enhance the capability of the model in capturing the underlying data structures. The widely used joint training paradigm in MvC is potentially not fully leverage the multi-view information, since the imbalanced and under-optimized view-specific features caused by the uniform learning objective for all views. For instance, particular views with more discriminative information could dominate the learning process in the joint training paradigm, leading to other views being under-optimized. To alleviate this issue, we first analyze the imbalanced phenomenon in the joint-training paradigm of multi-view clustering from the perspective of gradient descent for each view-specific feature extractor. Then, we propose a novel balanced multi-view clustering (BMvC) method, which introduces a view-specific contrastive regularization (VCR) to modulate the optimization of each view. Concretely, VCR preserves the sample similarities captured from the joint features and view-specific ones into the clustering distributions corresponding to view-specific features to enhance the learning process of view-specific feature extractors. Additionally, a theoretical analysis is provided to illustrate that VCR adaptively modulates the magnitudes of gradients for updating the parameters of view-specific feature extractors to achieve a balanced multi-view learning procedure. In such a manner, BMvC achieves a better trade-off between the exploitation of view-specific patterns and the exploration of view-invariance patterns to fully learn the multi-view information for the clustering task. Finally, a set of experiments are conducted to verify the superiority of the proposed method compared with state-of-the-art approaches on eight benchmark MvC datasets.

Figures (3)

• Figure 1: Illustration of imbalanced multi-view clustering issue in a two-view case. (a) The joint training multi-view clustering paradigm. (b) Our proposed balanced multi-view clustering approach. (c) The clustering performance of view-specific features extracted by diverse models. (d) The clustering performance of joint representation obtained by different models.
• Figure 2: Illustration of balanced multi-view clustering (BMvC). The multi-view data are first processed through their encoders to extract view-specific latent features. Next, a fusion module aggregates the representations into a unified joint one, which is then used to reconstruct the multi-view data via view-specific decoders. Finally, the view-specific contrastive regularization (VCR) and feature reconstruction loss are employed to guide the optimization of the model.
• Figure 3: The parameter sensitivity of the proposed method on eight benchmark multi-view datasets in terms of ACC, NMI, ARI, and Fscore, respectively.

Theorems & Definitions (2)

• Theorem 4.1
• proof