A CURVILINEAR GRADIENT PATH METHOD FOR OPTIMIZATION OF BIOLOGICAL SYSTEMS MODELS

摘要

We describe a novel curvilinear gradient method for least squares parameter estimation suitable for a wide class of biological systems models. At each iteration step, a minimum search is performed along a curvilinear trajectory corresponding to the steepest descent path of the least-squares objective function. The approach combines the benefits of both local (steepest-descent) and non-local (Newton-step) descent methods to offer a robust performance, especially suited to medium and large scale problems associated with biological models.