The proposed research will examine and assess a variety of parallel architectures for their capabilities on a range of perpceptual tasks. (1) A procedure will be developed to describe and specify multi- computers in terms of their basic components, assign costs, and from this assess hardware complexity and total costs; (2) Underlying topologies will be categorized, and generic, generalized structures and specific multi-computers chosen from each for further examination; (3) Algorithms will be developed (consulting architects, for balance) for a range of perceptual tasks, from image enhancement to object recognition and computer vision; (4) For each architecture, the performance (time and hardware needed) of each algorithm will be analyzed and estimated; for a selected set, the algorithm will be programmed and run. The results will give a systematic set of comparisons between topologies, and between particular multi-computers within each topology. Selected comparisons will show how closely actual performance is predicted by estimated performance, and indicate where and how to improve upon the specifications of basic components and their costs. Thus tools will be developed for relatively quickly estimating architectures' capabilities.
