Multi-level Reliable Guidance for Unpaired Multi-view Clustering

TL;DR

The work tackles unpaired multi-view clustering by introducing MRG-UMC, a three-module framework that learns consistent and confident cluster structures across views through inner-view multi-level clustering, synthesized-view alignment, and cross-view guidance. The method optimizes a joint objective with autoencoder representations and three module losses, and is supported by theoretical analysis showing reduced boundary errors and enhanced confidence via reliable views. Empirically, MRG-UMC delivers substantial improvements over state-of-the-art methods on five benchmarks, including notable gains in NMI, ACC, and F1, demonstrating its effectiveness for real-world, unpaired multi-view data. The approach offers scalable, cross-view clustering with principled confidence management, with potential impact on surveillance, multimedia, and cross-modal analytics.

Abstract

In this thesis, we address the challenging problem of unpaired multi-view clustering (UMC), which aims to achieve effective joint clustering using unpaired samples observed across multiple views. Traditional incomplete multi-view clustering (IMC) methods typically rely on paired samples to capture complementary information between views. However, such strategies become impractical in the UMC due to the absence of paired samples. Although some researchers have attempted to address this issue by preserving consistent cluster structures across views, effectively mining such consistency remains challenging when the cluster structures {with low confidence}. Therefore, we propose a novel method, Multi-level Reliable Guidance for UMC (MRG-UMC), which integrates multi-level clustering and reliable view guidance to learn consistent and confident cluster structures from three perspectives. Specifically, inner-view multi-level clustering exploits high-confidence sample pairs across different levels to reduce the impact of boundary samples, resulting in more confident cluster structures. Synthesized-view alignment leverages a synthesized-view to mitigate cross-view discrepancies and promote consistency. Cross-view guidance employs a reliable view guidance strategy to enhance the clustering confidence of poorly clustered views. These three modules are jointly optimized across multiple levels to achieve consistent and confident cluster structures. Furthermore, theoretical analyses verify the effectiveness of MRG-UMC in enhancing clustering confidence. Extensive experimental results show that MRG-UMC outperforms state-of-the-art UMC methods, achieving an average NMI improvement of 12.95\% on multi-view datasets. {The source code is available at: https://anonymous.4open.science/r/MRG-UMC-5E20.

Figures (7)

• Figure 1: The difference between single-level and multi-level clustering. In single-level clustering, some boundary samples typically emerge when there are $K$ clusters. In contrast, multi-level clustering forms clusters progressively from a coarse set $Cs = \{k_1, k_2, k_3\}$. For the cluster $k_3$, high-confidence sample pairs are mined across different levels with the help of $k_1$ and $k_2$, effectively reducing the number of boundary samples and resulting in a more confident cluster structure.
• Figure 2: The framework of Multi-level Reliable Guidance for UMC is built upon autoencoders with regularization. In latent spaces, training occurs in three stages with different numbers of clusters ($k_1$, $k_2$, and $k_3$). At each stage, three modules are employed: the inner-view multi-level clustering module, the synthesized-view alignment module, and the cross-view guidance module.
• Figure 3: The Venn diagram illustrates the sample sets in high-level and multi-level clustering, where $k_1 < k_2$. 'S' denotes the similar set, and 'D' denotes the dissimilar set.
• Figure 4: Evolution of the latent representation on the Digit dataset (6 views and 10 clusters) for Proposition 1. Each color denotes a cluster.
• Figure 5: Evolution of the latent representation on the Digit dataset (6 views and 10 clusters) for Proposition 2. Each color denotes a cluster.