Partitioning on Boltzmann machines

Different models for the solution of the partitioning problem on a Boltzmann machine (BM) are presented. Sequential and parallel BM optimizers are implemented and compared to existing heuristics. The results show that the sequential implementation can compete with classical heuristics, but in this case there is no advantage in time complexity and behavior compared to simulated annealing. The parallel implementation, which represents the advantage of the BM, is not applicable in this form for optimization, problems. Convergence cannot be guaranteed and will not be achieved for dense graphs close to full connectivity. Thus, a BM implementing the described dynamics is not directly applicable for general problems