A SQP-Method for Linearly Constrained Maximum Likelihood Problems

Autor:	Christian Kredler
Rok vydání:	1996
Předmět:	Generalized linear model Optimization problem Quadratic equation Estimation theory MathematicsofComputing_NUMERICALANALYSIS Applied mathematics Standard algorithms Convex function Nonlinear regression Mathematics Sequential quadratic programming
Zdroj:	Applied Mathematics and Parallel Computing ISBN: 9783642997914
DOI:	10.1007/978-3-642-99789-1_12
Popis:	Newton-like methods are standard algorithms for unrestricted parameter estimation in a wide class of nonlinear regression models. The search directions of the here presented algorithm are solutions from a sequence of quadratic (sub-)problems (SQP) with linear constraints. The practically important generalized linear models with natural link functions, e.g. log-linear or logistic regression, lead to strictly convex optimization problems for which this easy to implement extension of Newton’s method converges globally with quadratic rate. The numerical results are demonstrated at some ship damage data.
Databáze:	OpenAIRE
Externí odkaz:	https://explore.openaire.eu/search/publication?articleId=doi_________::40ca784bdf1e8f3dddde1c94672a4be1 https://doi.org/10.1007/978-3-642-99789-1_12 Zobrazit plný text záznamu