摘要
本文论述了前馈型网络,隐节点层进行线性独立变换的能力.严格证明了Mirchandani等关于隐节点在输入空间划分出区域数的定理.给出多层前馈型神经网络采用单位阶跃和连续渐近激发函数两种情况下,实数值样本绝对记忆能力的两个定理.
This paper mentions about the linearly independent transformation abilityof hidden layer in feed-forward nets. An exact, proof for Mirchandani's theorem which counts the regions of input space partitioned by the hidden nodes. Two theorems about real sample absolute recording ability corresponding with unit step and continuous asymptotic activation functions in multi-layer feed-forward neural networks, respectively, are provided.
出处
《计算机学报》
EI
CSCD
北大核心
1992年第7期536-540,共5页
Chinese Journal of Computers
基金
国家自然科学基金
关键词
神经网络
样本记忆能力
Neural networks, artificial neural networks, feed-forward nets, capacity of networks, sample recording ability.
