基于信赖域Newton算法的ELM网络  

ELM based on trust region Newton method

作　　者：韩敏[1] 王新迎[1] 

机构地区：[1]大连理工大学电子信息与电气工程学部,辽宁大连116024

出　　处：《控制与决策》2011年第5期757-760,共4页Control and Decision

基　　金：国家自然科学基金项目(60674073);国家科技支撑计划项目(2006BAB14B05);国家973计划项目(2006CB403405)

摘　　要：针对极端学习机(ELM)网络伪逆输出权值计算方法的运算复杂度制约其训练速度问题,提出一种基于信赖域Newton算法的新型ELM网络(TRON-ELM),并采用信赖域Newton算法求解ELM网络的输出权值.该算法首先构造一个ELM网络代价函数的Newton方程,并将其作为一个无约束优化问题,采用共轭梯度法求解,避免了求代价函数Hessian矩阵逆的运算,提高了训练速度,信赖域条件的存在保证了算法的整体收敛性.仿真实验结果验证了所提出方法的有效性.Considering the problems that the complexity of generalized inverse limits the learning speed of extreme machine learning（ELM）,a novel ELM,called TRON-ELM,is proposed based on the trust region Newton method in which the trust region Newton method is used to derive the output weights.The proposed method takes the Newton equation of the cost funcion of ELM as an unconstrained optimization,and a conjugate gradient method is used to solve the equation,which avoids solving the inverse of the Hessian matrix,thus the operation speed is improved.Meanwhile,the existence of trust region guarantees the global convergence.The experimental results show the effectiveness of the proposed method.

关 键 词：极端学习机 信赖域Newton法 共轭梯度法 回归 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]