Robust Low-Rank Graph Multi-View Clustering via Cauchy Norm Minimization

Graph-based multi-view clustering methods aim to explore the partition patterns by utilizing a similarity graph. However, many existing construct consensus graph based on original space, which may result in lack of information underlying low-dimensional space. Additionally, these often fail effectively handle noise present To address issues, novel graph-based method combines spectral embedding, non-convex low-rank approximation and processing into unit framework is proposed. In detail, proposed constructs tensor stacking inner product normalized embedding matrices obtained from each matrix. Then, decomposed tensor. The constrained via nonconvex Cauchy norm with an upper bound noise. Finally, we derive denoised experiments five datasets demonstrate that outperforms other state-of-the-art datasets.