Large-scale stochastic sparse subspace representation with consensus anchor guidance

Subspace clustering (SC) is a hotspot in data analysis and machine learning. There exists much literature addressing this topic and most of which cannot handle large scale data. Although anchor graph learning is introduced to SC, there is still a problem that anchors cannot preserve the subspace structure of original data and spectral clustering process is still implemented slowly. To address these issues, an Anchor Graph Regularization based Large-Scale Stochastic Sparse Subspace Representation with Consensus Anchor Guidance (AGLS\(^4\)RA) is proposed in this paper, which integrates three modules, including sparse self-representation, anchor graph regularization, and sparse coding into a unified framework. These modules are collaboratively worked to learn an optimal, high-quality anchor matrix under the row sparse constraint. Furthermore, the random sampling and label propagation techniques are also introduced to accelerate the clustering task. AGLS\(^4\)RA is capable of processing data in linear time, which is beneficial to the execution of large-scale tasks. A series of comparative experiments on benchmark datasets verify the effectiveness of the proposed method.