摘要
给出了一种只用单个隐元就可以解决Ｎ元奇偶校验问题的前向神经网络结构中的传递函数ｆ（ｘ）＝ｋｘ＋αＡπｓｉｎ２π２ｘ绕原点转动时函数单调非降性的必要条件：ｋ＞Ａ且０＜ｋ＜ｔｇπ２．
A feed forword neural net has been proposed,it can be used to N bits encoder problem with just one hidden unit which transformation function f(x) is a raised sinusoid.However the paper doesnt give the condition of monotone increasing characteristic of the f(x) when the f(x) rotates around the origin.The condition has been derived.
出处
《西南师范大学学报（自然科学版）》
CAS
CSCD
北大核心
1997年第3期261-264,共4页
Journal of Southwest China Normal University(Natural Science Edition)
基金
重庆市科委"九五"攻关项目
关键词
神经网络
奇偶校验
传递函数
单调非降性
neural network
odd even check
transformation function
monotone increasing characteristic
