论文部分内容阅读
本文提出了前馈神经网络学习的一种新理论——区间小波神经网络，不同于以往工作的是本工作的主要特点有：（１）采用区间小波空间作为神经网络的学习基底空间，克服了以往神经网络基空间与被学习信号所属空间不匹配问题；（２）由于采用区间小波理论，克服了原来被学习信号为适应神经网基空间而延拓所带来的不光滑性，使神经元数目得以节约，这在高维学习情形效果极为显著；（３）神经单元所用活性函数不再为同一个函数．
This paper presents a new theory of feedforward neural network learning - Interval wavelet neural network, different from the previous work is the main features of this work are: (1) Interval wavelet space as a neural network learning base space to overcome In the past, the basic space of neural network does not match the space of the signal to be learned. (2) Due to the use of the theory of interval wavelet, the original signal of the neural network is not smooth enough to adapt to the space of the neural network, The number can be saved, which is extremely significant in high-dimensional learning situations; (3) the activity functions used by neural units are no longer the same function.