An Algorithm for Least Squares Estimation of Parameters in Nonlinear Regression Models
暂无分享,去创建一个
The Gauss-Newton and Lavenberg-Marguardt methods are the most popular in the problem of parameters estimation in nonlinear models. These methods are based on the second order Taylor polynomials of the quadratic loss function. G-N method reply Hessian by Jacobian, while L-M method modifies Jacobian in the case when Hessian is not non negative defined. In the paper full information from Hessian is taken to define steps and directions in numerical procedure. Some simulations for regression functions of exponential and Tornguist type function are given. References:
[1] D K Smith,et al. Numerical Optimization , 2001, J. Oper. Res. Soc..
[2] T. M. Williams,et al. Practical Methods of Optimization. Vol. 1: Unconstrained Optimization , 1980 .
[3] D. Marquardt. An Algorithm for Least-Squares Estimation of Nonlinear Parameters , 1963 .