摘要
径向基函数网络(Radial Basis Function Network,RBFN)是二十世纪八十年代末提出的一种神经网络.当网络的输入维数较大时,RBFN的系统复杂性大大提高,从而使RBFN的行为受到影响,因此降低RBFN输入维数已成为RBFN的研究热点.本文提出一类基于RBFN的分工协作系统及其学习算法(A Divide-and-Cooperate Hybrid System Based RBFN,DCRBFN).DCRBFN是一种由多个子RBFN组成的混合结构,每个子RBFN具有自己的输入空间.由于DCRBFN把高维模型分解为低维模型,所以DCRBFN不仅明显降低了RBFN的复杂性而且网络的收敛速度更快.实验表明,DCRBFN在处理高维模型的行为明显优于RBFN.
In recent decade, due to simple architecture and learning, radial basis function networks(RBFN)have become one of the most popular models in neural networks. In general, the structural complexity of a RBFN depends on the number of the hidden nodes which is further related to the input dimension. This paper presents a divide-and-cooperate hybrid system based on RBFN and its learning algorithm. This new architecture is composed of several sub-RBFN, each of which takes a sub-input space as its input. The output of this new architecture is a linear combination of the sub-RBFNs'output with the linear coefficients acquired via learning together with each sub-RBFNs'parameters. Since this system divides a high-dimensional modeling problem into several low-dimensional ones, it can considerably reduce the structural complexity of a RBFN, where by the net's learning speed is improved. We have experimentally shown its outstanding performance on synthetic data in comparison with a conventional RBFN.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
2004年第3期380-384,共5页
Pattern Recognition and Artificial Intelligence
基金
国家自然科学基金(No.10371135)
关键词
分工协作系统
径向基函数网络
系统复杂性
Divide-and-Cooperate System
Radial Basis Function Networks
System Complexity
