Anchor-based Multi-view Subspace Clustering with Hierarchical Feature Descent

TL;DR

Anchor-based Multi-view Subspace Clustering with Hierarchical Feature Descent (MVSC-HFD) tackles view discrepancy and scalability in multi-view clustering by projecting heterogeneous views into a common subspace via hierarchical feature descent, learning a unified anchor space, and constructing a consensus bipartite graph for scalable clustering. The method employs alternating optimization to update hierarchical projections, a shared anchor matrix, and a nonnegative consensus graph with convergence guarantees, achieving a low-dimensional, joint embedding suitable for downstream clustering. Empirical results on 10 public datasets show MVSC-HFD consistently surpasses state-of-the-art MVC methods in ACC, NMI, and Purity while maintaining competitive runtimes and enabling linear-time performance on large-scale data. This framework offers a scalable, adaptable solution for multi-view data with varying dimensions and modalities, with practical implications for large-scale multimodal clustering tasks.

Abstract

Multi-view clustering has attracted growing attention owing to its capabilities of aggregating information from various sources and its promising horizons in public affairs. Up till now, many advanced approaches have been proposed in recent literature. However, there are several ongoing difficulties to be tackled. One common dilemma occurs while attempting to align the features of different views. {Moreover, due to the fact that many existing multi-view clustering algorithms stem from spectral clustering, this results to cubic time complexity w.r.t. the number of dataset. However, we propose Anchor-based Multi-view Subspace Clustering with Hierarchical Feature Descent(MVSC-HFD) to tackle the discrepancy among views through hierarchical feature descent and project to a common subspace( STAGE 1), which reveals dependency of different views. We further reduce the computational complexity to linear time cost through a unified sampling strategy in the common subspace( STAGE 2), followed by anchor-based subspace clustering to learn the bipartite graph collectively( STAGE 3). }Extensive experimental results on public benchmark datasets demonstrate that our proposed model consistently outperforms the state-of-the-art techniques.

Figures (6)

• Figure 1: The conventional NMF model applies a linear transformation to the original input space, which is usually of higher dimension. On the other hand, Deep NMF goes a step further by learning multiple layers of hidden representations that gradually reveal the ultimate lower-dimensional representation of the data in an hierarchical manner.
• Figure 2: The frameworks of the traditional methods and ours.
• Figure 3: The t-SNE plot of our proposed method on 3 benchmark datasets
• Figure 4: The relative running time of compared methods.
• Figure 5: The objective curve of our proposed method on 10 benchmark datasets