权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

Study of Modeling and Intelligent Control of Complex Systems Using Learning Networks

利用学习网络的复杂系统建模与智能控制研究

基本信息

批准号：
09450171
负责人：
HIRASAWA Kotaro
金额：
$ 6.08万
依托单位：
Kyushu University
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research (B)
财政年份：
1997
资助国家：
日本
起止时间：
1997 至 1999
项目状态：
已结题

项目摘要

Since the first proposal of a neuron model by Mc Culloch and Pitts in the 1940's, especially after the revitalization of artificial neural networks in 1980's, a variety of neural networks have been devised and are now applied in many fields. The vast majority of neural networks in use are those networks whose parameters or weights are tuned by gradiant-based supervised learning. This category includes feedfoward networks or multilayer parceptrons, various types of recurrent neural networks, radial basis function networks, fuzzy neural networks, and some networks with special architectures, such as time delay neural networks. These networks seemingly have different architectures and are trained by distinguishable training algorithms. In essence, however, they can be unified in a single framework in regard to both their architectures and learning algorithms. Universal Learning Networks (ULN's) have been proposed, as the name indicates, to provide a universal framework for the class of neural networks and moreover to model and control complex systems because most of the general complex systems in the real world can be modeled by the networks whose nodes represent the processing elements, and the branch between the nodes can describe the relation among the processes. Unification of a variety of network architectures which can describe the complex systems and unification of their learning algorithms are an objective of ULN's. This provides a consistent viewpoint for the various kinds of networks.

自20世纪40年代Mc Culloch和Pitts首次提出神经元模型以来，特别是80年代人工神经网络的复兴，各种各样的神经网络被设计出来并应用于许多领域。绝大多数使用的神经网络是那些参数或权重由基于梯度的监督学习调整的网络。该类别包括前向网络或多层Parceptron、各种类型的循环神经网络、径向基函数网络、模糊神经网络以及一些具有特殊架构的网络，例如时延神经网络。这些网络似乎具有不同的架构，并通过可区分的训练算法进行训练。然而，从本质上讲，它们可以在架构和学习算法方面统一在一个框架中。通用学习网络（Universal Learning Networks，简称ULN）的提出，顾名思义，是为了给神经网络类提供一个通用的框架，进而对复杂系统进行建模和控制，因为真实的世界中的大多数一般复杂系统都可以用网络来建模，网络的节点代表处理单元，节点之间的分支可以描述过程之间的关系。统一各种能够描述复杂系统的网络结构和统一它们的学习算法是ULN的一个目标。这为各种网络提供了一致的观点。