Parallel Local Search Algorithms

In general, neural networks are regarded as models for massively parallel computation. But very often, this parallelism is rather limited, especially when considering symmetric networks. For instance, Hoppeld networks do not really compute in parallel as their updating algorithm always requires sequential execution. We describe a recurrent network corresponding to a symmetric network and introduce a method of parallel updating multiple units. We show how this may be extended to Boltzmann machines with continuous activation functions, and point out possible applications of this architecture, e.g. local hill-climbing algorithms to solve the satissability problem. In terms of SAT-algorithms, we present an approach which allows the simultaneous change of truth value assignments for more than one propositional variable at a time, such that the theoretical properties of the considered algorithms are preserved, and give experimental evidence that this algorithm is indeed faster than the respective sequential variants.