摘要
根据认知的计算神经科学的观点,提出了一种基于神经系统动力学理论和连通图的信息的直接表达方式.它首先定义了知觉信息直接表达的神经结构和动力学模式,然后提出一个双层的网络计算模型,分别用于记录外界刺激的特征信息和连通对应的特定神经回路的连接模式,这是通过结构学习来实现的.在两层神经元间建立起来的连通结构同时起到联想记忆的作用,记忆的可靠程度由神经回路的连通度来决定.这种直接表达方式对于人工智能中关于语义表达和基于语义的推理研究具有重要意义.
One of the interferences between inheritance and concurrency is inheritance anomaly. From the view of cognitive computational neuroscience, a direct information representation method is presented based on neural system dynamics and graphic theory. A group of neurons and their connections representing perceptual information directly and the dynamical behaviors of neurons are defined firstly, and then a two-layer neural network is designed to record characteristics of stimulus and connect a specialized neural circuit that responding to the perception of that stimulus respectively. This could be achieved by the structure learning algorithm. The circuit constituted by neurons in two layers is also served as an associative memory of stimulus whose credibility is decided by the degree of connection of the circuit. The direct representation method is of very significance to the research of semantic representation and inference driven by semantics in artificial intelligence.
出处
《软件学报》
EI
CSCD
北大核心
2004年第11期1616-1628,共13页
Journal of Software
基金
国家自然科学基金
国家重点基础研究发展规划(973)
中国科学院智能信息处理重点实验室开放基金~~
关键词
联想记忆
表达
计算神经科学
动力学系统
associative memory
representation
computational neuroscience
dynamic system
