APPLICATION OF THE DARPA ADAPT PROCESSOR TO KNOWLEDGE-BASED EXPERT SYSTEMS

THE DARPA ADAPT PROCESSOR IS A NEURAL NETWORK SIMULATION THAT CAN IMPLEMENT MUTUAL INTERACTIONS BETWEEN A LARGE NUMBER OF PROCESSING ELEMENT VIA ADAPTIVE, PROGRAMMABLE WEIGHTING FUNCTIONS. THE PROCESSOR WAS DESIGNED PRIMARILY AS A MEANS TO SIMULTANEOUSLY CORRELATE MANY DIFFERENT TEMPLATES WITH A SET OF FEATURES EXTRACTED FROM DATA. IT WOULD SEEM, HOWEVER, THAT THE SAME ARCHITECTURE CAN BE USED FOR KNOWLEDGE-BASED EXPERT SYSTEMS (KBES'S). THE ADAPT PROCESSOR MAY BE EXTREMELY VALUABLE FOR AI APPLICATIONS, SINCE MANY DIFFERENT RELATIONS AND INFERENCES COULD BE IMPLEMENTED SIMULTANEOUSLY AT EACH TIME INCREMENT. THE NEED TO APPLY RULES TO A LARGE NUMBER OF INFERRED AND OBSERVED FACTS HAS SLOWED SOME KBES'S TO THE POINT WHERE THEY ARE INEFFECTIVE. THIS COMPUTATIONAL BOTTLENECK MAY BE AVOIDED THROUGH THE USE OF THE ADAPT PROCESSOR. PAY-OFFS INCLUDE SYSTEMS THAT CAN MAKE HYPOTHESES AS THEY GATHER DATA, AND THAT CAN CONTROL MULTISENSOR DATA GATHERING SO AS TO VERIFY OR REJECT THESE HYPOTHESES IN REAL TIME.