Simulated tempering and mean field annealing for mapping to multicomputers. (c1996)

Abstract

We worked on two physical optimization algorithms, Mean Field Annealing and Simulated Tempering, to solve the data mapping problem on multicomputers. We studied the effects of user defined parameters on MFA's behavior. These experiments resulted in recommending values to these parameters. Next, we implemented, for the first time, Simulated Tempering to solve the mapping problem. Then, we applied Mean Field Annealing, Simulated Tempering together with Simulated Annealing, Genetic Algorithm and Neural Network on randomly generated graphs. It was found that Neural Network (NN) proved to be the fastest algorithm although the quality of its solution is inferior to that of the Genetic Algorithm (GA), which was superior to all algorithms. Simulated Annealing had the second rank for both execution time and solution quality.