一种快速模糊ARTMAP神经网络算法  

A kind of Fast Fuzzy-ARTMAP Neural Network Algorithm

作　　者：葛辉[1] 

机构地区：[1]石河子广播电视大学,新疆石河子832000

出　　处：《微计算机信息》2007年第20期223-224,共2页Control & Automation

摘　　要：模糊ARTMAP神经网络学习算法可以动态高效地检测新闻文本中的模式,但随着网络大小的增加收敛速度变慢,计算复杂,鲁棒性不高。因此提出了一种快速模糊ARTMAP神经网络算法。该算法主要提出了新的匹配函数和激励函数,不需要规范化和补码计算步骤,去除了匹配跟踪过程,当发生未遂节点为赢得竞争的节点时,直接建立一个新的模板层的类标识。算法的革新之处在与既能动态维持聚类结构同时又减少了算法的复杂性。实验结果证实了该算法在合理时间内产生了高质量的模式发现。The Fuzzy-ARTMAP neural network algorithm can examine the pattern within the news text efficiently with the dynamic state. But the convergence velocity becomes slowly with the increase of the network dimension, the computation complexity and the Robustness anti higher. Therefore put forward a kind of fast fuzzy-ARTMAP neural network algorithm. That algorithm mainly put forward a new match and activation function which both simple for computation and understanding. Doing not need the normalization and complements computing the step, doing away with the match trace procedure, being to take place the uncompleted nodal point for win the nodal point of the competition, directly create a class identification of new template layer. The novelty of the proposed algorithm is the structure incrementally. Experimental results demonstrate that proposed clustering algorithm produces high-quality patterns discovery while fulfill a reasonable run time.

关 键 词：模糊自适应共振理论 神经网络 算法 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]