Research on Novel Clustering Algorithm Based on Convolutional Variational Auto-encoder
Yiran Fu · 2021
In this era of rapid technological development, the information that people come into contact with has increased exponentially. Every corner of human society has accumulated a variety of data. In this context. how to organize and utilize unlabeled data effectively becomes an urgent problem to be solved. This has promoted the development of a research hotspot in artificial intelligence, clustering. Recently, clustering algorithm based on auto-encoders have attracted widespread attention because of their excellent performance. However, when the data dimension is very high and its structure is relatively complex especially facing image data, the clustering method based on auto-encoder cannot get satisfactory clustering results. Meanwhile, these methods also have the disadvantage of not being able to directly learn the clustering results. To address the above issues this paper proposes a clustering algorithm based on the convolutional variational auto-encoder called Conv-VAE. Through the joint optimization learning of convolutional network and variational auto-encoder, the method can better process high-dimensional image data. The category-encoder embedded in the network can directly obtain the clustering results, avoiding complicated post-processing. Experiments on real image data sets demonstrate that our proposed algorithm is superior to related algorithms proposed in recent years in terms of clustering performance.