A continuous-time inference network for minimum-cost path problems

The proposed inference network requires no heuristics and uses essentially the same interconnection structure as described by K. Paham et al. (1990) to obtain all the minimum-cost paths simultaneously. Simple binary inferences at m-2 sites (for each of the C [m,2] units of the network) are required; and a conflict resolution scheme of applying local minimization with self-feedback is introduced. The dynamic behavior of the inference network is described by a set of C[m,2] parametrized differential equations, for which the convergence rate can be made arbitrarily fast and is practically independent of m. Numerical simulations using such a continuous-time inference network for 5-city and 10-city minimum-cost path problems are described and compared