论文部分内容阅读
提出了一种激励函数可调的新神经网络模型.和一般BP网络不同,在学习过程中,不但连接权值可以调节,神经元激励函数也可以调节,以双螺线问题为例,说明如何利用先验知识导出所需的激励函数,并给出网络的训练方法.实验表明,这种新型神经网络对于某些问题具有很强的分类能力.
A new neural network model with adjustable incentive function is proposed, which is different from general BP network. In the process of learning, not only the connection weights can be adjusted, but also the excitation function of neurons can be adjusted. Taking the double spiral problem as an example, Using the prior knowledge to derive the required incentive function, and give the training method of the network.Experiments show that this new neural network has strong classification ability for some problems.