CORC  > 自动化研究所  > 中国科学院自动化研究所  > 毕业生  > 博士学位论文
题名复杂性与联结机制网络的研究
作者刘晓鸿
学位类别工学博士
答辩日期1994-08-01
授予单位中国科学院自动化研究所
授予地点中国科学院自动化研究所
导师戴汝为
关键词复杂性 联结机制 反混沌 进化 自组织 混沌边缘 开放复杂巨系统 综合集成 综合集成研讨厅 开放巨型智能系统 人工生命 Hopfield网络 NK模型 随机布尔网络 元胞机 协同经济学 相变 量子计算机 环上的机器 超级Turing机 稀释 混沌 极限环 线性规划 椭球法 遗传算法 马蹄映射 异或 Complexity Connectionist Models Antichaos Evolution Self-Orgnization Edge of Chaos Open Complex Giant System Metasynthesis Hall
其他题名Complexity and Connectionist Models
学位专业模式识别与智能系统
中文摘要本论文的内容涉及复杂性和联结机制的有关内容.第一章是关于复杂性和开 放复杂巨系统的综合集成理论的一个综述.首先介绍了复杂性研究的基本模型和 工具,包括Hoptield网络、元胞机、Kauffman的随机布尔网络和NK模型、以及遗传 算法;在此基础上,比较详细地介绍了人工生命中混沌边缘的生命及向混沌边缘 的共同进化;接着讨论经济中的复杂性,介绍了沈华嵩的一个自组织经济模型, Zhang的协同经济学及其若干结论,以及Arthur的存在自增强机制即正反馈时经 济的有别于负反馈经济的一些特点,并简单介绍了考虑不完全理性时对系统进行 仿真的两种方案;Holland的分类系统和KauHman的经济网;之后,我们较为详细 地讨论了智能系统,包括对人工智能的历史回顾,对计算机不可以思维的若干代 表性论述,从而得到出路在于建立巨型智能系统,尤其是基于综合集成研讨厅理 论的开放巨型智能系统的结论。 后面是有关联结机制,尤其是Hopfield网络动力学的内容。第二章讨论了计算 模型,先给出了线性阈值单元所构成的神经网络与Turing等价的证明;接着,介绍 了两个很有影响的模型:Deutsch的量子计算机和Blum等人关于环上的机器的工 作,两者分别是由物理学家和数学家给出的,计算能力都较Turing机强;最后,对 一篇有关可以解决Turing机停机问题的“超级”Turing机的文章进行评述,指出了 其假设中的错误。 第三章为Hopfield网络上的极限环,先总结了网络的动力学性质,然后说明 可用极限环去表达概念,并尝试利用线性规划及遗传算法去建立网络.特别是,利 用哈奇扬的椭球法,可解决任意联结的网络的训练,从而彻底解决了网络的学习 问题,并给出若干实例说明其有效性,其中有例子是Hebb规则不能求得的。第四 章为Hopfield网络上的混沌,先给出了微分动力系统、迭代系统以及离散域上网络 的混沌的概念,说明了它们间的联系,之后,证明了对任意的阶数,存在Hopfield网 络有一个遍历所有点的极限环,接着讨论了“异或”问题,网络只有奇偶两个极限 环,对于阶数较低的情形,给出了其例子。 最后一章是全论文的总结,并指出了今后研究的方向.
英文摘要thesis is concerning on complexity and the connectionist models. The first chapter is a survey of complexity and the open complex giant system with its methodology, i. e., metasynthesis. We intrduce the basic models in complexity, whiche include the Hop field Network, the celluar automata, the random Boolean networks and the NK model of Kauffman. Then we introduce the origin of life at the edge of chaos and co--evolution on the edge of chos in length. Then we discuss complex phenomena in economy, which con sist of a self--organized model by Sheng, synegetic economics by Zhang with some of its conclusions, the self--reinforcing mechanism in economics by Arthur, and two simulation models under imperfect rationatity in economy, which are Holland's classffer system and Kauffman's economic web. Then we disscus the intelligent systems thoroughly, which in clude the history of AI ,and some typical resaons on why the computer can" t think, and the giant intelligent systems, espicially the giant open intelligent system based on the hall of the workshop of methasynthetic engineering. Then the connectionist models are disscussed, especiaUy the dynamics of Hopfield net works. The second chapter is about the computational models. We prove the equivalence of the linear threshold neural network and Turing machine. Then we introduce two models more powerful than Turing machine which are the quantum computer by Deutsch and the machine over the ring by Blum etc. on the view of the quantum physics and mathematics respectively. Then we point out the faults in the hypothesis of the Super -- Turing machine by Stannett which can solve the halting problem of Turing machine. The limit cycles on the Hopfield network are discussed in chapter 3. A summary of the behaviour of the discrete Hopfield network is given ,and representing concepts by the limit cycles is discussed. We setup the network by algorithms of linear programming and the genetic algorithms. With the ellipsoid method, we can train the network of any connections. by the given samples of either the fixed points and limit cycles or any sequences el. fectively. Some examples are given including examples that can't be solved by the Hebbrule. In chapter 4 we present the concept of chaos of the differential systems, the iterated systems and the networks baseed on the discrete fields. We prove that there exist chaotic Hopfield networks whose dynamical behaviour is a single limit cycle of all posible states for any order n. Then XOR problem is discussed ,and the cycles and the networks are giv. en for n smaller than 6. The last chapter is a concluding, remark, and the research directions in the near future.
语种中文
其他标识符294
内容类型学位论文
源URL[http://ir.ia.ac.cn/handle/173211/5644]  
专题毕业生_博士学位论文
推荐引用方式
GB/T 7714
刘晓鸿. 复杂性与联结机制网络的研究[D]. 中国科学院自动化研究所. 中国科学院自动化研究所. 1994.
个性服务
查看访问统计
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。


©版权所有 ©2017 CSpace - Powered by CSpace