摘要
将专家系统方法中类似医生凭化验单进行诊断的判断操作，归结为一个以n维超平行体为定性基准，以人工神经元为其定性基准一个（（n-1）维超平面）边界的定性映射．并指出，人工神经元可看作是一个定性基准仅有一个有限边界，而其余边界均消失在无穷远点的定性映射．若以一组人工神经元所围封闭邻域为其定性基准，则该定性映射等价于这组神经元构成的人工神经网络．还讨论了定性基准的伸缩、平移和叠加（或整合）等（线性）变换与人工神经元网络的关系，指出，定性基准的伸缩等价于连接权重的调节，平移等价于阈值调整，叠加等价于边界（人工神经元）旋转．也就是说，带定性基准线性变换的定性映射，不仅具有人工神经网络的所有调节功能，而且，可表示真值随定性基准而变的动态判断和识别过程．
It is shown in this paper that a judgment that such as doctor＇s diagnoses in expert system can be extracted as a dynamic Qualitative Mapping, whose criterion [α,β] is a n-dimension parallelepiped with a boundary of a （n-1） dimension hyperplane or an artificial neuron. An artificial neuron can be considered as a qualitative mapping whose criterion is made up of only one finite boundary, the rest boundaries all disappear at infinity. The relationship between some linear transformations of qualitative criterion and artificial neural network are respectively discussed. It is shown that flex transformation of criterion equivalent to the weight justification, shift transformation of criterion equivalent to the threshold modification, and the superposition transformation of differ dimension criterion equivalent to the rotation of the boundary （or artificial neuron）. It is said that the qualitative mapping with a linear transformation of criterion has all adjust functions of artificial network, besides, the dynamic judgment and recognition in which truth value varying with the qualitative criterion can also be described by the qualitative mapping.
出处
《哈尔滨工程大学学报》
EI
CAS
CSCD
北大核心
2006年第B07期6-12,共7页
Journal of Harbin Engineering University
基金
上海海事大学重点学科基金资助项目（XL0101-1）.
关键词
定性映射
定性基准
线性变换
转化程度函数
人工神经元网络
qualitative mapping
linear transformation
qualitative criterion
conversion degree function
fuzzy artificial neurons
