On the solution of the errors in variables problem using the l 1 norm

摘要: A fundamental problem in data analysis is that of fitting a given model to observed data. It commonly assumed only the dependent variable values are error, and least squares criterion often used fit model. When significant errors occur all variables, then an alternative approach which frequently suggested for this variables minimize sum squared orthogonal distances between each point curve described by equation. has long been recognized use not always satisfactory, thel1 superior when estimating true form contain some very inaccurate observations. In paper measure goodness taken be norm errors. Levenberg-Marquardt method proposed, main objective take full advantage structure subproblems so they can solved efficiently.