论文部分内容阅读
本文首先讨论了以M-P 神经元为标准的神经网络的特点及局限性。其次介绍了一种新的可用数字设备实现的概率逻辑神经元网络,比较了两者的特性。并指出前者在训练方法上和在硬件实施上的难点,强调了后者的易实现和训练方法的简单性。最后对后者网络状态的演变可用随机自动机理论进行分析的特点,显示出后者对于通用连接机分析的更一般理论。
This paper first discusses the characteristics and limitations of neural networks with M-P neurons as the standard. Secondly, we introduce a new probabilistic logical neural network that can be realized by digital equipment, and compare the characteristics of the two. And points out the difficulties of the former in training methods and hardware implementation, emphasizing the latter’s ease of implementation and training methods. Finally, the evolution of the latter state of the network can be analyzed using the stochastic automata theory, showing the latter’s more general theory of universal connector analysis.