论文部分内容阅读
以神经元局域场分布为基础,重新研究了连续神经元传输函数对具有联想记忆的人工神经网络功能的影响.与以往的认识不同的是,研究发现连续传输函数与硬极限传输函数相比并不存在明显的优越性,相反,连续传输函数对网络的某些功能,如最大存储率具有负面影响.研究表明神经网络的特性主要决定于网络的动力学结构(具体体现为网络吸引子对应的神经元局域场分布),网络的动力学结构可以通过选择合适的设计规则进行有效控制,不同的传输函数虽然也能影响到网络的动力学结构,但是它所带来的影响是被动的,可控性很差.
Based on the local distribution of neurons, the effect of continuous neuron transfer function on the function of artificial neural network with associative memory is re-studied. Unlike the previous knowledge, it is found that the continuous transfer function and hard limit transfer function On the contrary, the continuous transfer function has a negative impact on some functions of the network, such as the maximum storage rate.Research shows that the characteristics of neural networks mainly depend on the dynamic structure of the network (embodied as the network attractor corresponding , The dynamic structure of the network can be effectively controlled by selecting the appropriate design rules. Although different transfer functions can also affect the dynamic structure of the network, its impact is passive , Poor controllability.