Distributed implementation of optimization problems on bipartite network topology-fault tolerant computation and reduction of local minima problem

The author reports on advantages obtained by combined use of a bipartite network topology and a distributed implementation scheme for solving optimization problems. These properties are defective hardware, reduction of local minima, and parallel inter-unit communication. Several methods for deciding connection weights are provided. The probabilistic updating rule is illustrated by the implementation of the Boltzmann machine on bipartite network topology. A probabilistic updating scheme using analog units is given