BRST algorithm

Boender-Rinnooy-Stougie-Timmer algorithm (BRST) is an optimization algorithm suitable for finding global optimum of black box functions. In their paper Boender et al. describe their method as a stochastic method involving a combination of sampling, clustering and local search, terminating with a range of confidence intervals on the value of the global minimum. The algorithm of Boender et al. has been modified by Timmer. Timmer considered several clustering methods. Based on experiments a method named "multi level single linkage" was deemed most accurate. Csendes' algorithms are implementations of the algorithm of [Boender et al.] and originated the public domain software product GLOBAL. The local algorithms used are a random direction, linear search algorithm also used by Törn, and a quasi—Newton algorithm not using the derivative of the function. The results show the dependence of the result on the auxiliary local algorithm used.

Boender-Rinnooy-Stougie-Timmer algorithm (BRST) is an optimization algorithm suitable for finding global optimum of black box functions. In their paper Boender et al. describe their method as a stochastic method involving a combination of sampling, clustering and local search, terminating with a range of confidence intervals on the value of the global minimum. The algorithm of Boender et al. has been modified by Timmer. Timmer considered several clustering methods. Based on experiments a method named "multi level single linkage" was deemed most accurate. Csendes' algorithms are implementations of the algorithm of [Boender et al.] and originated the public domain software product GLOBAL. The local algorithms used are a random direction, linear search algorithm also used by Törn, and a quasi—Newton algorithm not using the derivative of the function. The results show the dependence of the result on the auxiliary local algorithm used. ==External links== http://www.abo.fi/~atorn/Globopt.html With the author's permission, text has been verbatim copied. Janka Compares various global optimization algorithms, of which BRST shows superior performance. Janka Presents the number of function-evaluations performed on the testset of Dixon-Szegö. Along with the MCS algorithm, the BRST requires the lowest number of evaluations.

Text derived from https://en.wikipedia.org/w/index.php?oldid=1329863190 under CC BY-SA 4.0; modifications have been made.