【摘要】 静态神经网络缺乏动力学行为,动态神经网络由微分方程描述,神经元具有反馈环,更适合描述动态系统.以前的研究中所提出的一些改进的递归网络中引入的都是输入层到关联层的反馈或是输出层到输入层的反馈,但是这些反馈系数是常数,是不可调的,限制了网络反映动态性能的能力.为此,提出了一种由带有积分器和可调反馈系数的神经元构成的动力学神经网络,并利用梯度下降法得到了网络的学习算法.更多还原

【Abstract】 The static neural network is short of dynamic behavior.Dynamic neural network is described by differential equations with feedback loop.It is more appropriate to descript the dynamic systems.The improved recurrent networks in the previous studies are introduced by the input-middle layer feedback or output-input feedback.These feedback coefficients are constant which limit the network to reflect the dynamic behavior.Therefore,this article puts forward to new dynamic neural networks with integrator and adjustable feedback coefficient and gets its learning algorithms by gradient descent.更多还原