Nonconvex optimization using a Fokker-Planck learning machine
Johan A. K. Suykens, Joos P. L. Vandewalle · 1995
A new algorithm for nonconvex optimization by means of a so-called Fokker-Planck Learning Machine is proposed in this paper. This is done by considering the Fokker-Planck (FP) equation related to continuous simulated annealing, which has been proven to convergence to the global optimum under certain conditions. An approximate solution to the FP equation is sought by parametrizing the transition density by means of Gaussian sum approximations (Radial Basis Function networks). Like in genetic algorithms a population of points is considered. At each generation the points are generated based upon the transition density and the density is updated according to the FP equation.