没有合适的资源?快使用搜索试试~ 我知道了~
首页张神经网络的有限时间收敛及其在实时矩阵平方根求解中的应用
张神经网络的有限时间收敛及其在实时矩阵平方根求解中的应用
0 下载量 51 浏览量
更新于2024-08-29
收藏 488KB PDF 举报
本文探讨了一种新型的有限时间收敛张量神经网络(Finite-Time Convergent Zhang Neural Network,FTCZNN),并将其应用于实时矩阵平方根求解。相比于原始的张量神经网络(Original Zhang Neural Network,OZNN)模型,FTCZNN在设计上引入了非线性激活函数——双幂函数,这显著提高了网络的收敛速度。该文的核心贡献在于,作者理论推导并确定了FTCZNN模型的收敛时间上界,通过解相关的微分不等式来实现这一估计。 在矩阵平方根问题中,传统的数值方法可能存在计算复杂度高、收敛速度慢的问题。FTCZNN的优势在于其能够在较短的时间内找到矩阵的精确或近似平方根,这对于实时计算和在线优化任务具有重要意义。文章首先回顾了张量神经网络的基本原理,强调了它们在解决非线性问题上的潜力,然后重点介绍了FTCZNN的设计原理、特性以及与OZNN模型的区别。 在实验部分,通过模拟比较,在相同的条件设置下,FTCZNN模型显示出更快的收敛速度和更优的性能。这不仅验证了新模型的有效性,也证明了它在实际应用中的优越性,特别是在处理大规模和密集型矩阵时,能够提供实时且高效的解决方案。 关键词包括:张量神经网络、矩阵平方根、有限时间收敛、非线性激活函数、收敛时间上界。这篇文章对神经网络理论的拓展和优化算法的实际应用结合紧密,对于提高计算效率,特别是在实时计算领域,具有重要的理论价值和实践意义。
资源详情
资源推荐
Neural Comput & Applic
DOI 10.1007/s00521-017-3010-z
ORIGINAL ARTICLE
A finite-time convergent Zhang neural network
and its application to real-time matrix
square root finding
Lin Xiao
1
Received: 7 January 2016 / Accepted: 11 April 2017
© The Natural Computing Applications Forum 2017
Abstract In this paper, a finite-time convergent Zhang neu-
ral network (ZNN) is proposed and studied for matrix square
root finding. Compared to the original ZNN (OZNN) model,
the finite-time convergent ZNN (FTCZNN) model fully uti-
lizes a nonlinearly activated sign-bi-power function, and
thus possesses faster convergence ability. In addition, the
upper bound of convergence time for the FTCZNN model
is theoretically derived and estimated by solving differential
inequalities. Simulative comparisons are further conducted
between the OZNN model and the FTCZNN model under
the same conditions. The results validate the effectiveness
and superiority of the FTCZNN model for matrix square
root finding.
Keywords Zhang neural networks · Matrix square root ·
Finite-time convergence · Nonlinear activation function ·
Upper bound
1 Introduction
The solving problem of matrix square root has been the sub-
ject of much research in recent years and widely applied
in various scientific and engineering areas [1–4]. Much
effort has been directed towards the matrix square root find-
ing [4–9] because of its usefulness as a tool in computers.
In mathematics, in order to solve the matrix square root
Lin Xiao
xiaolin@jsu.edu.cn; xiaolin860728@163.com
1
College of Information Science and Engineering,
Jishou University, Jishou 416000, China
problem, almost all algorithms/schemes are based on the
following defining equation [1–9]:
X
2
(t) − A = 0, (1)
where coefficient matrix A ∈ R
n×n
is assumed to be known
(or at least measurable accurately). If A ∈ R
n×n
has no non-
positive real eigenvalues, then there is a unique solution X,
which is denoted by A
1/2
and called the principal square
root of A [1–4]. For presentation convenience, let X
∗
denote
the theoretical principal square root of matrix A,which
corresponds to the actual X(t) solved by various methods.
In this paper, we assume that A has no non-positive real
eigenvalues and we are interested in the computation of the
principal square root.
In general, there are two classes of methods for solving
the matrix square root problem. One is based on the numer-
ical algorithms performed in a serial manner. The other is
based on the parallel-processing approaches (e.g., neural
networks). As a typical iterative numerical algorithm, New-
ton iteration [2] has good properties of convergence and
stability and has been investigated and applied to matrix
square root finding. After that, many improved algorithms
from Newton iteration have further been developed and ana-
lyzed for solving matrix square root problems [2–4], such as
simple form of Newton iteration [2] the Meini iteration [3],
and the Denman and Beavers iteration [4]. However, these
numerical algorithms may encounter serious speed bottle-
neck due to the serial nature of the digital computer, and
may not be efficient enough for large-scale online appli-
cations. As another important class of solution approach,
many parallel-processing computational schemes have been
developed, investigated, and implemented on specific archi-
tectures [10, 11] because of the parallel distributed nature.
Especially, as a software and hardware implementable
下载后可阅读完整内容,剩余7页未读,立即下载
weixin_38678510
- 粉丝: 8
- 资源: 967
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
最新资源
- 李兴华Java基础教程:从入门到精通
- U盘与硬盘启动安装教程:从菜鸟到专家
- C++面试宝典:动态内存管理与继承解析
- C++ STL源码深度解析:专家级剖析与关键技术
- C/C++调用DOS命令实战指南
- 神经网络补偿的多传感器航迹融合技术
- GIS中的大地坐标系与椭球体解析
- 海思Hi3515 H.264编解码处理器用户手册
- Oracle基础练习题与解答
- 谷歌地球3D建筑筛选新流程详解
- CFO与CIO携手:数据管理与企业增值的战略
- Eclipse IDE基础教程:从入门到精通
- Shell脚本专家宝典:全面学习与资源指南
- Tomcat安装指南:附带JDK配置步骤
- NA3003A电子水准仪数据格式解析与转换研究
- 自动化专业英语词汇精华:必备术语集锦
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功