论文部分内容阅读
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.
A novel ELM network based on trust region Newton algorithm (TRON-ELM) is proposed to solve the problem of its training speed, which is caused by the computational complexity of computing method of pseudo-inverse output weights of extreme learning machine (ELM) networks. The trust region Newton algorithm ELM network.The algorithm constructs a Newton equation of cost function of ELM network as a unconstrained optimization problem by using the conjugate gradient method to avoid the inverse operation of the cost function Hessian matrix and improves The existence of training speed and trust region guarantees the overall convergence of the algorithm.The simulation results verify the effectiveness of the proposed method.