论文部分内容阅读
本文提出了一种连续型前馈回归网络用于进行模拟值样本的联想记忆,网络的权值通过BP算法学习得到.用这种方法可以保证每个待记样本在可预先设计的凸域内是吸引的,并给出了一种球形吸引域的设计方法.为加深理解,文中给出了此种回归网与全连接动态网(Hopfield网)权值间的关系.
In this paper, a continuous feedforward regression network is proposed for associative memory of analog samples, and the weights of the network are obtained by the BP algorithm. In this way, it can be guaranteed that each sample to be memorized in the pre-designed convex domain is Attracting, and gives a design method of spherical attracting domain.In order to deepen the understanding, the relationship between the weight of this kind of regression net and the total connected dynamic network (Hopfield network) is given.