论文部分内容阅读
提出了一种具有自适应类警戒参数的模糊ＡＲＴＭＡＰ神经网络，为不同的模糊ＡＲＴ的类族设置了不同的警戒测试参数，并在学习过程中进行自适应调整。还提出了新的非交叠超方形以及非交叠的Ｎｅｓｔｅｄ超方形的建立与扩展学习规则。新的神经网络模型可以提高识别率，解决了发生在传统模糊ＡＲＴＭＡＰ神经网络中的记忆稳定性和弹性问题，并解决了传统的模糊ＡＲＴＭＡＰ神经网络不能处理的非凸输入特征空间的分类问题。
A fuzzy ARTMAP neural network with adaptive alert parameters is proposed. Different alert test parameters are set for different fuzzy ART families, and adaptive adjustment is made during the learning process. A new non-overlapping hyper-square and non-overlapping Nested hyper-square learning rules are also proposed. The new neural network model can improve the recognition rate and solve the memory stability and elasticity problems that occur in the traditional fuzzy ARTMAP neural network and solve the classification problem of the non-convex input feature space which can not be processed by the traditional fuzzy ARTMAP neural network.