Stochastic approximation techniques, applied to parameter estimation in a biological model

Simultaneous perturbation stochastic approximation (SPSA) is a class of optimization algorithms which compute an approximation of the gradient and/or the Hessian of the objective function by varying, all the elements of the parameter vector simultaneously and therefore, require only a few objective function evaluations to obtain first or second-order information. Consequently, these algorithms are particularly well suited to problems involving a large number of design parameters. In this study, their potentialities are assessed in the context of nonlinear system identification. To this end, a challenging modeling application is. considered, i.e. dynamic modeling of batch animal cell cultures from sets of experimental data. The performance of the optimization algorithms are discussed in terms of efficiency, accuracy and ease of use.