Multi-layer Topology Preserving Mapping for K-Means Clustering

摘要

In this paper, we investigate the multi-layer topology preserving mapping for K-means. We present a Multi-layer Topology Preserving Mapping (MTPM) based on the idea of deep architectures. We demonstrate that the MTPM output can be used to discover the number of clusters for K-means and initialize the prototypes of .K-means more reasonably. Also, K-means clusters the data based on the discovered underlying structure of the data by the MTPM. The standard wine data set is used to test our algorithm. We finally analyse a real biological data set with no prior clustering information available.