global optimization; stochastic algorithms; competing heuristics; non-linear regression; NIST datasets; MATLAB

The paper deals with algorithms for estimation of non-linear regression parameters. Stochastic population-based algorithm with competition was implemented and compared with standard gradient algorithm commonly used for least-squares estimates. The results show that this stochastic algorithm found the global minimum in most tasks where gradient algorithm fails. Such population-based algorithms can be used as a tool for estimation of non-linear regression parameters, especially in tasks of higher difficulty level or in tasks when suitable starting values for gradient method are not available.

The paper deals with algorithms for estimation of non-linear regression parameters. Stochastic population-based algorithm with competition was implemented and compared with standard gradient algorithm commonly used for least-squares estimates. The results show that this stochastic algorithm found the global minimum in most tasks where gradient algorithm fails. Such population-based algorithms can be used as a tool for estimation of non-linear regression parameters, especially in tasks of higher difficulty level or in tasks when suitable starting values for gradient method are not available.