前馈神经网络WAFS学习算法

WAFS Learning Algorithm for Feedforward Neural Networks

1.引言前馈神经网络是目前应用最广的一种神经网络,其学习算法是由Rumelhart等人于1986年提出的反向传播(Back Propagation,BP)算法,故这种神经网络也常被称为BP神经网络。人们对前馈神经网络学习算法的研究,以前主要着重对各层之间联接权值优化的研究。

A new learning algorithm named WAFS (Weight, Activation Function and Structure) for feedforward neural networks is proposed in this paper. Unlike other methods that have been published, this method takes advantages the three main factors of neural network,i. e. ,weights ,activation functions and topological structure. This method is based on parametric neural networks and structural learning, and learning rate matrix is applied to it. The simulation results of two-spiral problem and WBCD(Wisconsin Breast Cancer Diagnosis)problem have proved the success of this method.