结合遗传算法的聚类神经网络

neural Networks Combining Genetic Algorithms for Clustering

许多聚类算法有两个缺点：１）采用某种距离作为相似性测度。类别接受域为球形，不能与复杂模式分布匹配；２）对确定合理类别数不能提供任何帮助。采用最大似然准则的聚类算法其类别接受域为球形或椭球形，可以与模式的分布匹配更好。在计算似然值时使用先验概率，能为确定合理的类别数提供依据。本文的贡献是把遗传算法结合到基于最大似然准则的神经网络聚类算法中，解决聚类中心的初值选择问题并获得最优聚类。

Many clustering algorithms have two problems:1)the receptive fields of clusters are spherical which can not match the vwrious distributes of patterns because of using distance metric;2)they do not provide any clue regarding the number of clusters.Using maximum likelihood criterion in clustering algorithms,the clusters have spherical or ellipsoidal receptive fields and match the distributes of patterns better.It provide a indication to determine the best number of clusters that the priori probabilities of clusters are used in calculating likelihood values.The contribution of this paper is that genetic algorithms are combined with clustering neural networks based on maximum likelihood criterion.That solves the problem of selecting the initial cluster centers and results in the best clustering.