关键词: 聚类, 类中心, 能量模型, 概率, 极值

Abstract: Clustering is a complex issue. Although there is a variety of clustering methods, many shortcomings still exist, such as slow clustering convergence, unsatisfactory clustering results, requiring certain parameters provided by people. To solve these problems, a new idea of clustering was put forward. Firstly, the authors supposed each cluster had a cluster center. Secondly, each sample point was considered as an energy source, eradiating energy to the clustering space with a reasonable physical or mathematical diffusing model. Cluster center was confirmed by the total energy that each point gained. Finally, as a result, sample points could be easily clustered to their cluster centers. The experimental results demonstrate that this clustering approach has the characteristics of fast convergence, strong extendibility, and being suitable for natural clustering. Additionally, it can obtain the same results of many classic clustering methods.

Key words: clustering, cluster center, energy model, probability, extremum