摘要
本文在仔细分析神经网络知识存储方式的基础上，指出了现有存储方式的严重不足：由大量具有简单处理能力的神经元组成的神经网络，虽然具有一定的智能处理能力，但由于每个神经元不具备复杂的处理能力，故必然导致由此构成的网络存在着诸如局部极小、收敛速度缓慢、推广能力差等缺点，尤其是难于用于实时处理系统，大大限制了神经网络的应用范围。为此，本文提出了一种新的智能型神经元模型并将它与常用的神经元模型进行了比较，后续论文“高阶神经网络及广义知识存储原理”
Based on in-depth analysis of knowledge storing mode of neural networks,the inadequacy of current storing mode is pinpointed.That is to say,despite its definite intelligent processing capability.the neural neiwork made up of large numbers of neurons with simple processing capability but without complicated processing capability is bound to have such drawbacks as local minimum,slow converging rate and insufficient capability of being popularized.In particular,it can hardly be applied to real-time processing systenis,which greatly limits its range of application.To remedy such a situation,a new intelligent neuron model is proposed and compared with neuron models in common use.It is pointed out in a subsequent paper entitled 'Generalized Information Storing Principle and Higher-Order Generalized Neural Networks' that the algorithm based on the intelligent neuron model can raise the converging rate of the conventional BP algorithm more than 1,000 times.
出处
《电子学报》
EI
CAS
CSCD
北大核心
1996年第4期86-90,共5页
Acta Electronica Sinica
基金
武汉青年晨光计划资助
中国博士后科学基金
关键词
神经网络
学习算法
网络模型
BP算法
Neural networks,Learning algorithm,Network model,BP algorithm
