Orthogonal Non-negative Tensor Factorization based Multi-view Clustering

A.2 Proof of the 2nd part From Weierstrass-Bolzano theorem, there exists at least one accumulation point of the sequence P Inspired by [2], we adopt directly alternate sampling (DAS) to select anchors. Specifically, if the features are negative, we process the features of each dimension by subtracting the minimum value in each dimension. C.1 Experimental Configurations The Reuters and NoisyMNIST are implemented on a standard Windows 10 Server with two Intel (R) Xeon (R) Gold 6230 CPUs 2.1 GHz and 128 GB RAM, MATLAB R2020a. The MSRC, HandW ritten4, Mnist4 and AW A are implemented on a laptop computer with an Inter Core i5-8300H CPU and 16 GB RAM, using Matlab R2018b. We repeated the all methods 20 times independently and showed the averages with the corresponding standard deviations. The specific hype-parameters on each dataset are as follows: MSRC: anchor rate = 0.7, p = 0.5, λ = 100.