An adaptive randomized pattern search

Abstract
The problem of functional minimization by means of random searching is discussed. Two previously proposed algorithms, Adaptive Step Size Random Search (ASSRS) and Randomized Pattern Search (RANPAT), are evaluated on several functions which were chosen so as to be difficult if not impossible to minimize using simple gradient search techniques. The experimental evaluation of the ASSRS and RANPAT algorithms leads to the development of a proposed hybrid random search algorithm, Adaptive Randomized Pattern Search (ARPS). Comparisons of these three algorithms indicates that ARPS is at least comparable to ASSRS and RANPAT on all the functions examined, and that ARPS is significantly better on the several functions which have constrained parameter spaces.

This publication has 6 references indexed in Scilit: