Alternative simulation annealing processes for global optimizationin neural networks

摘要

Summary form only given, as follows. The simulated annealingnmethod is a tool for finding the global minima of a performance measurenfunction. It is accomplished by constraining the probabilityndistribution of the process to be a Gibbs distribution associated withnthe measure to be minimized. The only parameter upon which thenconvergence depends is the cooling schedule of the Gibbs temperature.nAlternative distributions have been derived along with the coolingnschedules for convergence to a global minimum. A global measure ofnperformance was defined. It was concluded that an algorithm using simplenmultiplicative and additive functions will perform faster on a computernthan an algorithm using more complicated functions