A parallel algorithm for Boltzmann machines (Q1191047)

The Boltzmann machine is a neural network model for which the author suggests a new synchronously parallel algorithm. The sequential Boltzmann machine calculates the Gibbs distribution of the energy function over the configuration space whence we can solve the optimization problem of the energy function by generally using the simulated annealing technique. But this last algorithm does not exploit the parallelism of the Boltzmann machine. The proposed algorithms are independent of the connection pattern and ensure that the equilibrium distributions are the Gibbs distributions and thus guarantee that they solve the optimization problems. The method is being tested by computer simulations. Reviewer's remark: I suggest to compare it with a new deterministic global optimization method (called Alienor) allowing to obtain good approximations of global minima.