PARALLEL PROCESSING FOR ARTIFICIAL INTELLIGENCE (AI) APPLICATIONS

KNOWLEDGE-BASED SYSTEMS PROVIDE THE POTENTIAL TO DEVELOP NEW CAPABILITIES FOR AVIONICS SOFTWARE SYSTEMS. KNOWLEDGE-BASED SYSTEMS CAN INTELLIGENTLY ADAPT TO THEIR ENVIRONMENT, FOCUS RESOURCES ON INTERESTING EVENTS, FILTER AND SUMMARIZE COMPLEX INPUT DATA, AND CAN BE EASILY AND RAPIDLY MODIFIED TO ACCOUNT FOR CHANGES IN ENEMY TACTICS OR OTHER ASPECTS OF THE TACTICAL ENVIRONMENT. THESE BENEFITS CAN ONLY BE ACHIEVED IF RULE-BASED SYSTEMS ARE MADE OF THE PRACTICAL. FOR AVIONICS APPLICATIONS, THIS "PRACTICALITY" LITERALLY TRANSLATE INTO SPEED. TO ACHIEVE REAL-TIME OPERATION OF THESE TYPES OF SYSTEMS, ONE MUST DEVELOP TECHNIQUES FOR "PARALLELIZING" WHAT WE INTRINSICALLY SEQUENTIAL INFERENCING PROCESSES, AND FOR IMPLEMENTING THESE TECHNIQUES ON PARALLEL PROCESSING ARCHITECTURES. THE PURPOSE OF THE PROPOSED WORK IS TWO-FOLD: (1) TO IDENTIFY SUCH "PARALLELIZING" TECHNIQUES AND OTHER TECHNIQUES FOR ACHIEVING REAL-TIME AI FOR AVIONICS; AND (2) TO SELECT A PREFERRED PARALLEL ARCHITECTURE THAT HAS GREATEST POTENTIAL FOR ACHIEVING THE DESIRED SPEEDS. PHASE II WILL FOLLOW BY ACTUALLY IMPLEMENTING THESE CONCEPTS IN THE CONTEXT OF A REAL-TIME DEMONSTRATION.