摘要
在简要描述三层前馈神经网络模型中最佳平方逼近输出向量与输出误差的数学表示的基础上 ,讨论了三层前馈神经网络的泛化性能 ,推导出与学习模式有偏差的微扰模式应满足的条件和异联想实现条件 。
In this paper, the generalizing function of feed forward three layered is discussed on the basis of a brief introduction to the mathematical description of least square approximative output vector and error vector. The mathematical restriction on the deference between training pattern and the perturbational pattern is derived, and the implementation condition of heteroassociation is also discussed. Finally, a heteroassociation algorithm for feed forward three layered neural networks is described.
出处
《延边大学学报(自然科学版)》
CAS
2001年第4期263-267,共5页
Journal of Yanbian University(Natural Science Edition)
关键词
三层前馈神经网络
泛化性能
微扰模式
异联想
算法
Feed forward three layered neural networks
Generalizing function
Perturbational pattern
Heteroassociation