A RT 2神经网络的一种改进

传统A RT 2神经网络在聚类过程中模式的匹配度量仅仅与模式的相位信息相关，这种匹配度量忽略了模式的幅度信息的作用，在对相位信息相同而幅度信息不同的两个簇进行聚类时，效果很差；同时，它还存在输入域限制的问题。针对这些不足之处，提出了一种改进的A RT 2神经网络，在输入模式进入网络学习过程中，保存其幅值信息，放宽对负实数的非线性转换，并考虑输入模式到各个簇的中心点的最短距离，同时增加一个阈值对离群点进行判定，消除了离群点对聚类结果的影响。实验验证，改进的A RT 2网络在对相同相位的两个簇聚类时，性能明显优于传统的A RT 2网络。

Improvement for ART2 neural network

While the matching measure of the pattern in clustering is only about the phase information and neglects the effects of the amplitude information of the patterns, traditional ART2 neural network can not cluster well for two clusters with the same phase but different amplitudes, and it also has limitation problem for inputs domain. As to the above disad-vantages, an improved ART2 algorithm is put forward. The amplitude information of the patterns is also saved during the input pattern entering the network in the learning process, and the limitation is relaxed in a nonlinear transformation of negative numbers, and the shortest distance from the input pattern to the center of each cluster is taken into consideration. At the same time, a threshold to judge outliers is added to eliminate the influence of outliers on clustering results. Experi-mental results demonstrate that the performance of the improved ART2 is superior to the traditional ART2 when they clus-ter the two clusters with the same phase.