论文部分内容阅读
设计了带参数的单极性和双极性柔性Sigmoid函数的一种柔性前馈神经网络(FBPN),并给出了相应的学习算法.和普通的前馈神经网络(BPN)不同,FBPN不仅能学习连接权,且同时能学习柔性Sigmoid函数的参数,因此,它能根据学习样本集,为每一个隐含层和输出层单元产生合适的Sigmoid函数形态.一个算例和二个应用实例说明,柔性神经网络能提高BP网络的性能,并能较好解决不同领域中的分类与预测问题.
A flexible feedforward neural network (FBPN) with unipolar and bipolar flexible Sigmoid functions is designed and a corresponding learning algorithm is given. Unlike ordinary feedforward neural networks (BPN), FBPN not only Can learn the connection right, and at the same time can learn the parameters of the flexible Sigmoid function, so it can generate proper Sigmoid function form for each hidden layer and output layer unit according to the learning sample set.An example and two application examples illustrate , Flexible neural network can improve the performance of BP network and can better solve the classification and prediction problems in different fields.