自增强的深度子空间聚类

In recent years, with the rapid development of the information age, a large amount of data has been generated on the Internet. In order to fully mine the information in the data, cluster analysis is a very effective means.The neural manifold clustering and embedding algorithm uses maximal coding rate reduction to extract features. On this basis, this paper imposes a constraint on the output of the classifier, and proposes a self-enhanced deep subspace clustering algorithm. Specifically, this paper first use maximal coding rate reduction to extract the features of the data, and then use the learned features to train the classifier. This paper also sharpen the output of the classifier to obtain a new probability distribution, and the KL divergence constraint or square loss constraints are used to reduce the distance between these two probability distributions, making the output of the classifier more specific, additionally this paper compare the impact of different sharpening factors on the experimental performance. This paper conducts experiments on CIFAR-10, CIFAR-100 and STL-10 datasets, and compares it with other clustering algorithms. This paper also conduct ablation experiments on each module in the model, and finally proves the effectiveness of this algorithm.