The Finnish Journal of Business Economics 4/1974, 329-336.
This paper discusses a gradient method which is well suited for
fitting nonlinear regression models by least squares minimization.
A way is suggested to speed up the convergence by applying
exponential smoothing for the gradient. The paper also discusses
adjustments of steplength in order to gain momentum in the
minimization and in order to damp out unnecessary oscillation, and
the use of scaling-factors in special cases.
It is clear that such cases can be devised where the minimization
procedure discussed does not work well enough. However, according to
the authors' computational experience it seems that in cases of
nonlinear regression analysis normally arising the method presented
is highly satisfactory.
Keywords: optimization methods, gradient methods, nonlinear
regression analysis
[To the fulltext paper ]