Deep Embedding Clustering Based on Residual Autoencoder

Clustering algorithm isone of the most widely used and influential analysis techniques.With the advent of deep learning, deep embedding clusteringalgorithms have rapidly evolved and yield promising results. Much ofthe success of these algorithms depends on the potential expressioncaptured by the autoencoder network. Therefore, the quality of thepotential expression directly determines the algorithm’sperformance. In view of this, researchers have proposed manyimprovements. Although the performance has been slightly improved,they all have one shortcoming, that is, too much emphasis is placedon the original data reconstruction ability during the process offeature expression, which greatly limits the further expression ofpotential features according to specific clustering tasks. Moreover,there is a large amount of noise in the original data, so blindlyemphasizing reconstruction will only backfire. Hence, weinnovatively propose a deep embedding clustering algorithm based onresidual autoencoder (DECRA) after in-depth research. Specifically,a novel autoencoder network with residual structure is proposed andintroduced into deep embedded clustering tasks. The networkintroduces an adaptive weight layer in feature representation z,which can make it have good robustness, generalization for specifictasks, and adaptive learning of better feature embeddings accordingto category classification. In this paper, the reasons for thevalidity of this structure are explained theoretically, andcomprehensive experiments on six benchmark datasets includingvarious types show that the clustering performance of the DECRA isvery competitive and significantly superior to the most advancedmethods.