摘要
从首一无平方多项式F(x,y,t)有根x=Фi(y,t)i=1,2,…,degx(F),其中,Ф(y,t)=Ci,0(t)+Ci,1(t)y+Ci,2(t)y2十……,入手,设计出一类含参二元多项式求根的神经网络模型和多项式近似因式分解的神经网络模型,为研究其神经网络学习算法,给出基于代数神经网络的含参多元多项式近似团式分解的基本理论。探讨基于代数神经网络的二元多项式的近似因式分解算法,提出一种新的确定误差代价函数的学习方法,相比梯度下降学习法,不存在局部极小问题,通过该算法学习,能逼近给定二元多项式F(x,y,t)的不可约因式,实现其对F(X,y,t)的因式分解。
ms paper, we use the fact tha monic squarefree contain parameter multuvariate polynomia F (x,y, t) has the roots x=φi (y, t), i=1,2,'', degx. (F), of the form Fi (y, t) =Ci0 (t)+Ci1(t) y+Ci2(t)Y2+..., akinds of nearel networks model of approximate factorizaion was designed. In orde to study learning algorithm,the basic theory of algebra neural networks of multivariate polynomial approximate factorization was given.We discussed contain parameter multivariate polynomials approximate factorization leaming algorithm. Anew kind of decision error price function learning algorithm, no exist utmost small problem, we use the algrothm learning, we get divisors. Finally we have realizated factorization of F (x, y, t).
出处
《计算机工程与设计》
CSCD
北大核心
2000年第3期58-63,共6页
Computer Engineering and Design
关键词
多元多项式
神经网络
近似因式分解
代数
multivarite polynomials
factorization
neural networks
approximate factorization
learning algorithm