Pseudo-Supervised Deep Subspace Clustering

Auto-Encoder (AE)-based deep subspace clustering (DSC) methods have achieved impressive performance due to the powerful representation extracted using neural networks while prioritizing categorical separability. However, self-reconstruction loss of an AE ignores rich useful relation information and might lead indiscriminative representation, which inevitably degrades performance. It is also challenging learn high-level similarity without feeding semantic labels. Another unsolved problem facing DSC huge memory cost n×n matrix, incurred by self-expression layer between encoder decoder. To tackle these problems, we use pairwise weigh reconstruction capture local structure information, a learned layer. Pseudo-graphs pseudo-labels, allow benefiting from uncertain knowledge acquired during network training, are further employed supervise learning. Joint learning iterative training facilitate obtain overall optimal solution. Extensive experiments on benchmark datasets demonstrate superiority our approach. By combining with k-nearest neighbors algorithm, show that method can address large-scale out-of-sample problems. The source code available: https://github.com/sckangz/SelfsupervisedSC.