Parallel distributed computation in vision

Describes a method of distributing many separate processes over a number of processors that arises out of work in neural networks and computer vision. The method results from the use of ADAM, an advanced distributed associative memory (see Austin, 1987). In conventional parallel computer systems each separate process (or computation) to be performed is undertaken on individual processors at given time intervals. At any moment in time one processor will be only computing one process. The method of time slicing between processes has been used for many years, and has the effect of maximizing the utilization of the processor and allowing processes to get the same amount of computation time over a period of time. The difficulty with this approach is that the scheduling of the processes is difficult because the time complexity of the algorithms are often not known. The author describes how a neural network processor being developed at York University for vision problems is being used to investigate parallel distributed computation (PDC). In this approach each process to be computed is not assigned to individual processes or even given time slots to run in. All processes run concurrently and on all processors. Thus no scheduling is required