忆阻神经系统的动力学演化与控制设计

项目名称： 忆阻神经系统的动力学演化与控制设计

项目编号： No.61304057

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 吴爱龙

作者单位： 湖北师范学院

项目金额： 23万元

中文摘要： 经典的计算机系统无法突破传统存储体系中难以解决的存储墙问题。由大规模忆阻纵横闩构建成的忆阻神经系统能很好地模拟类脑结构的学习和记忆等行为，使得忆阻神经系统更能接近人脑。该系统的内部动力学性态与存储模式问题是紧密相关、有共性的。这样，忆阻神经系统的动力学演化和控制设计将成为新的研究热点。本项目将构建基于忆阻仿生的集成混合电路，在一定程度上模拟人脑学习模式，充分考虑系统的混合逻辑动态和渐近行为特性，发展基于忆阻的神经系统的非线性动力学理论方法与新技术。结合忆阻突触的深度学习优势，初步建立不同时间尺度下的突触演化的动力学过程与基于忆阻仿生逻辑电路的存储模式之间的深层结构规律，实现信息通过系统簇的连接权被动态地存储起来，为基于忆阻的认知存储和获取提供强有力的理论指导。本课题的研究，将对混杂系统、协调控制、电子信息系统等等的研究产生一定的推进作用。

中文关键词： 忆阻神经系统；混杂控制；模式流；联想记忆；

英文摘要： Classical computer systems can not breach the memory wall problem, which is difficult to resolve in the conventional storage structures. Memristive neural systems consisting of large-scale memristor crossbar latches are capable of simulating neuromorphic cognitive behaviours such as learning and memory, moreover,the electronic intelligence systems can mimic the awesome power of human brains. Dynamical behaviors of these systems and the issues in stored patterns are close relatives. Therefore, dynamic evolution and control design of memristive neural systems will become a new research hotspots field. In this project, we will establish different classes of hybrid integrated circuits based on memristors, in a way that mimic the learning patterns of human brains. Considering fully the mixed logic dynamics and asymptotic behaviors of such systems, the nonlinear dynamic theory, methods and new techniques based on memristive neural systems will be developed. Combining with memristive deep learning advantages, the deep structure rules between dynamic processes evoked by memristive synapses in different time scales and storage patterns based on memristive biomimetic circuits will be studied, to achieve the dynamic memory of informations via the connection weights of network cluster, which might provide some strong theore

英文关键词： Memristive Neural Systems；Hybrid Control；Pattern Flows；Associative Memory；