DESIGN OF A MASSIVELY PARALLEL SIMD ARCHITECTURE FOR THE BOLTZMANN MACHINE

The Boltzmann Machine represent an interesting computation paradigm that can be applied in several fields such combinatorial optimization, pattern recognition and knowledge representation. Although Boltzmann Machine has been proved to give high quality results, its application in real problems has been limited by the heavy computational load it requires mainly due to the sequential nature of the simulated annealing algorithm needed for equilibrating the network. Thank to a parallel algorithm, we have developed, Boltzmann Machine features can now be exploited by using a parallel machine. We describe the design of a first prototype of a parallel SIMD architecture that supports the Boltzmann Machine and allows a speedup of N where N is the number of the processing nodes