Recently, it has been shown that representing the multi-view data by a tensor and then learning a latent self-expressive tensor is effective. However, early works mainly focus on learning essential ...