层次双曲函数型网络的整体逼近理论及学习算法

Universal Approximate Theory and Learning Algorithm of Hierarchical Hyperboloid Function Link Aritifical Neural Networks

1 引言函数型网络FLANN(Function Link ArtificalNN)最早由鲍·约翰提出,其实质是将输入经函数变换展成为一组基函数,而输出则为基函数的线性组合,然后,再根据实际需要选取基函数,基函数可用三角函数,多项式函数等。由于输出是基函数的线性组合。

In this paper,a kind or function link artifical neural networks is designed and a new hierarchical function link artifical neural networks HFLANN is presented,and its universal approximate property is disscussed,the universal learning algorithm is presented. HFLANN is very suitable for application domains with hierarchical knowledge structures,and therefore,it has very important practical significance.