基于模糊协方差的自适应聚类神经网络

将模糊协方差距离测度引入到竞争学习型神经网络的参数控制中，采用批处理的网络学习方式消除数据样本顺序对网络权重调整的影响，通过淘汰及合并数据集的冗余类实现对未知类别数、多种分布型数据的自适应聚类．实验表明，新网络对数据集的分布形式有较强的鲁棒性，并能正确确定数据集的类别数．

Adaptive Clustering Neural Network Based on Fuzzy Covariance Matrix

Fuzzy covariance matrix is introduced into the parameters controlling of competitive learning neural network. Batch processing network-learning mode is adopted to eliminate the influence of input data order on network-weight learning. Adaptive clustering for the data sets with unknown number of clusters and varied distribution is realized during rejecting and merging redundant clusters. The experiments show that the new network has strong robustness to the distribution of data sets and can obtain the number of clusters in the data correctly.