Artificial neural networks for distributed adaptive routing on dynamic topology networks

In considering distributed adaptive routing schemes for large networks with dynamic topology, the need for an unconventional shortest path algorithm arises from the excessive computation overhead associated with repeated path/distance calculations. The authors provide the design specifics of such an algorithm, and establish its performance characteristics through rigorous analysis and simulation. The new algorithm exploits the intrinsic parallelism of neural network architectures and solves the single-pair shortest path problem in such a way that: (i) the computation time is independent of the number of network nodes; and (ii) the frequent shortest distance/path re-calculations inherently associated with topology changes are performed much faster than conventional algorithms. Simple exploitation of the inherent parallelism allows extension of the algorithm to solving single-source and all-pair shortest path problems without compromising the trait of constant convergent complexity