Clark C. Guest
Assoc Professor, Electrical and Computer Engineering
Neural networks, goal-directed AI, optical system design, computervision Professor Guest is a researcher of artificially intelligent systems, fiber-optic communications, and computervision. Much work in AI aims at duplicating subliminal human processing, including speech and facial recognition. In this domain, Guest is working with a graduate student to develop a Chinese speech-recognition system. But another of Guest's interests is goal-directed artificially intelligent systems handling challenges to which humans must consciously apply analytical skill. A test case is an engine for playing strategic games. Such a model would be more efficient than existing machines that use brute-force parsing of all available options. In addition to military planning, a goal-directed AI model could assist humans in many fields, including business decision-makers. Guest's group developed the CAN neural network algorithm, which was orders of magnitude faster than the then prevalent back-propagation method. Guest also is an optical device expert, and his focus is on interconnect technology. Optics for this application is gaining new attention because of light's speed, low power, and other advantages that are at a premium at a time when computer-system architects are finding it increasingly difficult to power, cool, communicate, and operate devices running at ever higher clock rates. Guest's group was the first to provide a practical demonstration of the superiority of optical interconnections and was a leading developer of simulated annealing in the design of computer-generated holograms. Guest's work in computervision has been applied commercially to difficult production-line inspection tasks. In 1994, his group was the first to successfully apply the Neocognitron vision model to grayscale images and binocular depth perception.
Jacobs School Faculty Update Your Profile