AN IMPROVED MARQUARDT PROCEDURE FOR NONLINEAR REGRESSIONS

F.B. Smith, Jr. and D.F. Shanno, 1971, Technometrics, 13:63-74

Abstract           Go Back

A procedure for solving nonlinear estimation problems, designed with the special case of fitting nonlinear regression relationships in mind, will be described. It is an analytic search technique using two classic search procedures in parametric combination. This enables one at each iteration to concentrate attention on minimizing sum of squares deviation with respect to the single combination parameter. Using projected gradients, the procedure may also be applied to linearly constrained problems.