Enhanced Latent Multi-view Subspace Clustering
Long Shi, Lei Cao, Jun Wang, Badong Chen
TL;DR
This work tackles latent multi-view subspace clustering by identifying the limitation of simple view concatenation in preserving inter-view consistency. It proposes ELMSC, which constructs an augmented data matrix X_a with block diagonal stacking for complementary information and off-diagonal blocks X(p,q) that encode inter-view similarity via cosine similarity S(p,q) computed on PCA reduced features, and jointly learns an augmented latent representation H_a and augmented self-representation Z_a using an ADMM-based optimization with robust errors. Theoretical connections to existing LMSC are established and convergence to a stationary point is proven; extensive experiments on six real-world datasets show that ELMSC outperforms state-of-the-art baselines and exhibits robustness to parameter settings. The approach yields clearer affinity structures, stable performance across parameter choices, and competitive running times, highlighting its practical impact for scalable, high-quality multi-view clustering in real applications.
Abstract
Latent multi-view subspace clustering has been demonstrated to have desirable clustering performance. However, the original latent representation method vertically concatenates the data matrices from multiple views into a single matrix along the direction of dimensionality to recover the latent representation matrix, which may result in an incomplete information recovery. To fully recover the latent space representation, we in this paper propose an Enhanced Latent Multi-view Subspace Clustering (ELMSC) method. The ELMSC method involves constructing an augmented data matrix that enhances the representation of multi-view data. Specifically, we stack the data matrices from various views into the block-diagonal locations of the augmented matrix to exploit the complementary information. Meanwhile, the non-block-diagonal entries are composed based on the similarity between different views to capture the consistent information. In addition, we enforce a sparse regularization for the non-diagonal blocks of the augmented self-representation matrix to avoid redundant calculations of consistency information. Finally, a novel iterative algorithm based on the framework of Alternating Direction Method of Multipliers (ADMM) is developed to solve the optimization problem for ELMSC. Extensive experiments on real-world datasets demonstrate that our proposed ELMSC is able to achieve higher clustering performance than some state-of-art multi-view clustering methods.