一个大脑学习与记忆的现实性神经网络模型

A Realistic Neural Network on Cerebral Learning-memory

现实性神经网络是相对人工神经网络而言的 ,它与人工神经网络相比 ,更接近真实的生物神经网络。大脑学习与记忆的功能及特点 ,一直是许多学科研究的热门话题。建立一个关于大脑学习与记忆的现实性神经网络模型 ,将为研究大脑学习与记忆的功能、特点提供十分有用的工具 ,并有助于进一步理解大脑高级功能的机制。针对大脑学习与记忆具有的特点 ,着重考虑了 3个方面的问题 :1)如何使模型既具有稳定性又具有可塑性 ;2 )如何使模型的学习与记忆具有条件性 ;3)如何使模型学习与记忆具有一定随机性。通过数学分析及模拟实验表明 ,模型能较好地反映大脑学习与记忆的主要特点。如神经联系的条件性 (强化则巩固 ,不强化则衰减 ) ,大脑活动的随机性等 ,并简略地反映了DNA -RNA -Protein系统对记忆的意义。

Compared with artificial neural networks, realistic neural networks are much closer to real biological neural networks. The features and functions of the cerebral learning and memory are always research problems in great demand to many branches of science. Establishing a realistic neural network model on cerebral learning memory will offer a useful tool used to research the features and functions of the cerebral learning and memory, and contribute to understand the higher functional mechanism of human brain.This article has set up a realistic neural network on cerebral learning memory. In this model, to counter the features of the cerebral learning and memory, the following problems were mainly considered: ①How to make the model possess stability and plasticity; ②How to make learning and memory of the model possess conditionality;③How to make learning and memory of the model possess some randomness. The results of the mathematics analyses and experiments show that the model can reflect well the main features of the cerebral learning memory, such as the conditionality of neural connection (reinforcing, then consolidating;not reinforcing, then declining), the randomness of the brain activity and so on. And the significance of DNA RNA Protein system for memory is briefly reflected.