摘要
利用前向网络输入元素非线性关联的方法实现了从输入模式空间到输出标识空间复杂的非线性变换.推导了学习方法,并在学习过程中把模式的平移不变识别、比例不变识别及旋转不变识别等条件构造在神经网络的权结构之中,使其具有模式不变识别能力,同时借助等权类的概念,极大地简化了网络的拓扑结构。
Complicated nonlinear transformations are realized with the feedforward neural network, of which the input elements are connected nonlinearly. The learning rules of the network are derived, and the conditions of translation invariance, scale invariance and rotation invariance pattern recognition are constructed in the network during the learning procedure.Furthermore, with the help of the idea of equivalent weight subset proposed, the network is made much simple and the learning time is reduced.
出处
《北京邮电大学学报》
EI
CAS
CSCD
北大核心
1997年第2期49-54,共6页
Journal of Beijing University of Posts and Telecommunications
基金
邮电部科研基金
关键词
神经网络
非线性关联
学习方法
感知器
模式识别
neural networks
nonlinear connection
learning rule
invariant pattern recognition