A Novel Robust Low-rank Multi-view Diversity Optimization Model with Adaptive-Weighting Based Manifold Learning

Highlights：

• For more robust decomposition of the noisy part of the data, the L21 norm and the nuclear norm are used to constrain the error matrix of NMF and the error matrix of basis matrix, respectively.

• In order to preserve the geometric structure and relevant information of each view, adaptive-weighting manifold learning and the Hilbert Schmidt Independence Criterion are added to the model, which is solved by the idea of adaptive exponential weighting.

• The proposed algorithms have obtained very good experimental results in several well-known multi-view datasets, and it has a very fast convergence rate.

摘要

•We propose a novel multi-view clustering algorithm which features sparse low-rank subspace with the novel bilinear error matrices decomposition model based on non-negative matrix factorization (NMF) and adaptive-weighting manifold learning.•For more robust decomposition of the noisy part of the data, the L21 norm and the nuclear norm are used to constrain the error matrix of NMF and the error matrix of basis matrix, respectively.•In order to preserve the geometric structure and relevant information of each view, adaptive-weighting manifold learning and the Hilbert Schmidt Independence Criterion are added to the model, which is solved by the idea of adaptive exponential weighting.•The proposed algorithms have obtained very good experimental results in several well-known multi-view datasets, and it has a very fast convergence rate.

论文关键词：Low-rank Representation (LRR),Multi-view Subspace Clustering (MVSC),Hilbert Schmidt Independence Criterion (HSIC),Non-negative Matrix Factorization (NMF),Adaptive-Weighting Manifold Learning (AWML)