摘要
证明了区间小波神经网络具有一致及Ｌ２逼近性质，且为相容的函数估计子，其学习收敛速度在ｄ维情形不随ｄ增大而减慢，本质上克服了神经网络高维学习的“维数灾难”问题，模拟实例验证了理论的正确性．
In the present paper, it is proved that the interval wavelets neural networks has universal and L 2 approximation properties and is a consistent function estimator. Convergence rates associated with these properties do not decrease as d increases in d dimensional function learning, i.e. , the “curse of dimensionality” is eliminated substantially. In the experiments, the proposed interval wavelet neural networks, compared to traditional wavelet networks, has performed better.
出处
《软件学报》
EI
CSCD
北大核心
1998年第4期246-250,共5页
Journal of Software
基金
国家自然科学基金
国家863高科技项目基金
国家攀登计划基金
关键词
神经网络
小波
多尺度分析收敛
Neural network, wavelets, multiresolution analysis, convergence.
