Robust and fast subspace representation learning for multi-view subspace clustering

Abstract

Multi-view subspace clustering (MVSC) plays an indispensable role in the domains of data mining and machine learning. Compared to single-view analysis, this integration of information leads to more accurate and comprehensive clustering results, providing a solution for large-scale data clustering. Notably, various techniques have been proposed in the field. In the present context, most multi-view clustering methods mainly focus on enhancing the consistency of clustering and handling noise. Adapting multi-view subspace clustering effectively for the clustering of big data poses a significant challenge. To overcome this challenge, we propose a new method called “robust and fast subspace representation learning for multi-view subspace clustering (RFSR)”, which utilizes a unified encoder to process information from each view and integrates the information between different views. In this process, we reduce the impact of noise, employing either correntropy or ${\ell }_{\mathrm{2,1}}$-norm for handling it. Specifically, we start by randomly sampling from each view and then process the sampled data for noise. Subsequently, we train a unified encoder for each view to leverage complementary information from multiple views, thereby enhancing the robustness of clustering. We not only consider the multi-view data features but also account for its large scale and noise structure. Furthermore, we demonstrate through experiments the efficiency and robustness of our approach in multi-view subspace clustering.