CONCURRENT NEUROMORPHIC ALGORITHMS FOR OPTIMIZATION OF THE COMPUTATIONAL LOAD OF A HYPERCUBE SUPERCOMPUTER.

A novel approach to the optimization of the time-dependent computational load in message-passing multiprocessor systems is presented. Compact neuromorphic data structures are used to model effects such as precedence constraints, processor idling time, and task-schedule overlap. Analytic expressions are given for the effect of single-neuron perturbations on the systems' configuration energy. Algorithms for the implementation of this methodology on a hypercube supercomputer are outlined along with potental extensions to large-scale nonlinear asynchronous neural networks.