EASING THE CONSCIENCE OF THE GUILTY NET

This article deals with dynamical system neural networks which find fo r given vertices simple paths joining the vertices with optimal (short est) or near-optimal length. Such neural networks might have a role in scheduling a machine which wires integrated circuits. We develop a mo dified guilty net algorithm which is fully parallel in vertex placemen t; thus there might be efficient, albeit not perfect, implementations in parallel. During iterations of the algorithm all vertices of a math ematical path move in two-dimensional space until vertex positions coi ncide with given physical vertices.