Anchor Learning with Potential Cluster Constraints for Multi-view Clustering

TL;DR

ALPC tackles the challenge of scalable, high-quality anchors in multi-view clustering by introducing a shared latent semantic constraint that forces anchors to be generated from multiple data clusters and by aligning anchor graphs with the data’s clustering structure. The method jointly learns view-specific anchors and a consistent anchor graph, using variables $A^{(v)}$, $Z$, $P$, $R$, and $U^{(v)}$, and enforces orthogonality and nonnegativity to ensure discriminative representation. Through an alternating optimization with closed-form updates, ALPC achieves state-of-the-art clustering performance on six benchmarks, with favorable convergence and linear scalability in the number of samples. This approach offers practical benefits for large-scale, multi-view data by producing more representative anchors and compact, informative graphs that reflect the underlying clustering layout.

Abstract

Anchor-based multi-view clustering (MVC) has received extensive attention due to its efficient performance. Existing methods only focus on how to dynamically learn anchors from the original data and simultaneously construct anchor graphs describing the relationships between samples and perform clustering, while ignoring the reality of anchors, i.e., high-quality anchors should be generated uniformly from different clusters of data rather than scattered outside the clusters. To deal with this problem, we propose a noval method termed Anchor Learning with Potential Cluster Constraints for Multi-view Clustering (ALPC) method. Specifically, ALPC first establishes a shared latent semantic module to constrain anchors to be generated from specific clusters, and subsequently, ALPC improves the representativeness and discriminability of anchors by adapting the anchor graph to capture the common clustering center of mass from samples and anchors, respectively. Finally, ALPC combines anchor learning and graph construction into a unified framework for collaborative learning and mutual optimization to improve the clustering performance. Extensive experiments demonstrate the effectiveness of our proposed method compared to some state-of-the-art MVC methods. Our source code is available at https://github.com/whbdmu/ALPC.

Figures (6)

• Figure 1: Schematic diagram of the proposed ALPC. Aims to learn the most representative anchors for large multi-view datasets. ALPC unifies anchor learning and consistent anchor graphs into a single model. ALPC adopts the semantic constraint of sharing potential cluster structures with the aim of guiding the anchors to be generated uniformly from different clusters of the original data. At the same time, in order to ensure the representative nature of the generated anchors, ALPC discretizes the anchor graphs to guide the center of mass clustering of the original data and the mass clustering centers of the anchors are aligned, thus constraining both to share the same clustering structure.
• Figure 2: Traditional dynamic generation of anchors (left) and our proposed cluster-shared generation of anchors (right)
• Figure 3: (a) and (b) are sensitivity analyses of hyperparameters $\lambda_1$ and $\lambda_2$ to the test benchmark dataset, while (c) and (d) are sensitivity analyses of the number of anchors $m$ on two different datasets.
• Figure 4: The visualization of the complete graph of ALPC on both datasets BBCSport and MSRC.
• Figure 5: Convergence curves of ALPC on two datasets MNIST and Wiki.