Abstract :
An attempt was made to develop a quickly converging, connectionist, shortest-path algorithm based on simple update rules for a digital environment. An analog connectionist, shortest-path algorithm proposed by R. Marcus (1987) was determined to be inappropriate for digital implementation. This conclusion was based on the algorithm´s sensitivities to path length and step size, sensitivities which, in many cases, slow the convergence. A directional search algorithm which expands on Marcus´ analog representation and solves these shortcomings by incorporating a search criterion based on the application of relaxation methods is presented. The proposed method combines an arc activating with deactivating search methodology based on relaxation methods to obtain self-adapting step sizes. This approach clearly provides much faster convergence in digital applications and, as such, makes the connectionist representation viable for practical applications
