Low-rank representation coefficients regularized by the tensor-Singular Value Decom-position(t-SVD)scheme for multi-view subspace clustering have achieved impressive performances.However,all of them suffer from the following two common demerits.(1)They focus on exploring the relationships among samples to construct representations which are then stacked to be a tensor,whose computational complexity is at least O(n2 logn);(2)They always deploy the standard spectral clustering algorithm directly on the integrated representation,neglecting the prior knowledge of different representations towards the final results.To tackle these problems,we propose a novel Tensor Learning Induced Multi-view Spectral Clustering(TLIMSC)approach,where the spatial cluster structures and complementary information are simultaneously explored.Specifically,multi-view spectral embedding representations related from samples to clusters are focused to be stacked in a tensor,where the complexity finally becomes O(nlogn).Later,a bridge would be built to connect the learned representations carrying different adaptive confidences with the final consensus results.Extensive experiments on five datasets reveal the effectiveness and efficiency of TLIMSC.
NETL
NSTL
万方数据
