Robust regression computation computation using iteratively reweighted least squares

摘要: Several variants of Newton’s method are used to obtain estimates solution vectors and residual for the linear model $Ax = b + e b_{true} $ using an iteratively reweighted least squares criterion, which tends diminish influence outliers compared with standard criterion. Algorithms appropriate dense sparse matrices presented. Solving system updated matrix factorizations or (unpreconditioned) conjugate gradient iteration gives most effective algorithms. Four weighting functions compared, results given well-conditioned ill-conditioned problems.