Deep Multi-View Clustering Based on Reconstructed Self-Expressive Matrix

Deep Multi-view Subspace Clustering is a powerful unsupervised learning technique for clustering multi-view data, which has achieved significant attention during recent decades. However, most current methods rely on self-expressive layers to obtain the ultimate results, where size of matrix increases quadratically with number input data points, making it difficult handle large-scale datasets. Moreover, since multiple views are rich in information, both consistency and specificity images need be considered. To solve these problems, we propose novel deep approach based reconstructed (DCRSM). We use reconstruction module approximate coefficients using only small training samples, while conventional model must train network entire also shared specific integrate consistent information features fuse between views. The proposed DCRSM extensively evaluated datasets, including Fashion-MNIST, COIL-20, COIL-100, YTF. experimental results demonstrate its superiority over several existing methods, achieving an improvement 1.94% 4.2% accuracy maximum 4.5% NMI across different Our yields competitive even when trained by 50% samples whole