多层前馈神经网络的面向对象的程序设计框架

An Object-Oriented Programming Framework for Designing Multilayer Feedforward Neural Network

提出多层前馈神经网络的一种面向对象的程序设计框架,其中包括表示前馈神经网络中的神经元、权值的数据成员,以及完成网络的构造、初始化、学习和输出值计算等功能的成员方法,并用支持面向对象的程序设计语言Java实现了这一框架。模式分类和函数近似等问题的解决说明了这一前馈神经网络的面向对象的程序设计框架的有效性。

In this paper, an object-oriented programming framework for designing multilayer feedforward neural network is presented. The framework comprises the data members representing neurons and weights, and the member methods for performing construction, initialization and training of network, and output computation. Then, the framework is implemented employing Java programming language supporting object-oriented programming. By the use of the framework implemented, some problems of pattern classification and function approximation are solved. The results show the effectiveness of the framework.