Parallel Implementations Of Simulated Annealing
A local timing model for parallel optimization with Boltzmann Machines
It is well-known that synchronous fully parallel Boltzmann Machines do not asymptotically converge to an optimal solution in many optimization problems. Therefore the level of parallelism has to be reduced by an application specific partition of simultaneous active processing units. In this paper we present a local random timing model which automatically splits up the network in runtime controlled by a single global parameter. The speeding up of solving an optimization problem is discussed at the example of the TSP-problem.
Unable to display preview. Download preview PDF.
- Hinton, G.E., Sejnowski, T.J., Ackley, D.H.: "Boltzmann machines: constraint satisfaction network that learn", Carnegie-Mellon University (USA) Technical Report CMU-CS-84-119, (1984).Google Scholar
- Zwietering, P.J., Aarts, E.H.L.: "The Convergence of Parallel Boltzmann Machines", in Parallel Processing in Neural Systems and Computers, Editors: G. Eckmiller, G. Hartman, G. Hauske, North-Holland, pp. 277–280, (1990).Google Scholar
- Kreuzer, I., Goser, K.: "A Modified Model of Boltzmann Machines for WSI-Realization", EUSIPCO '88, Signal Processing IV, pp. 327–330, (1988).Google Scholar
- Aarts, E.H.L., Korst, J.H.M.: "Simulated Annealing and Boltzmann machines", Wiley Chichester, (1989).Google Scholar