AN ADAPTIVE NEURAL NETWORK ANTENNA CONTROLLER WITH PERFECT MEMORY AND ONE-SHOT LEARNING

CURRENT PHASED-ARRAY CONTROL TECHNIQUES ARE UNABLE TO ADAPT TO ANTENNA DEGRADATIONS AND FAILURES TO PRODUCE AN OPTIMIZED RADIATION PATTERN. ADAPTIVE CONTROL WILL IAINIMIZE UNPREDICTED FAILURES# REDUCE MAINTENANCE COATS# SLID INCREASE MACHINE AVAILABILITY. DEVELOSOFT PROPOSES TO DESIGN# DEVELOP# AND TEST A NEURAL NETWORK ANTERMA CONTROLLER IGHICH WILL CONTINUE TO PRODUCE AIL OPTIIAAL OUTPUT RADIATION PATTERN IN SEVERELY DEGRADED CONDITIONS. WE WILL USE PERFECT BEHAVIOR IAEIAORY TO COIAPARE ACTUAL MITH EXPECTED OUTPNT RADIATION. NOVEL ALGORITHMS WILL UPDATE THE ADAPTIVE NETWORK TBROUGH ONE-SHOT LEARNING THAT ELIIAINATES ACCURACY AND CONVERGENCE PROBLEMS FOUND IN CONVENTIONAL GRADIENT DESCENT TECHNIQUES. WE WILL DEMONGTRATE THE FEASIBILITY OF PERFECT MEMORY ADAPTIVE CONTROLLERS AS WELL AS THE EFFECTIVENESS OF ONE-SHOT UPDATE. OUR RESEARCHERS EXCEL IN THE FIELD OF NEURAL NETWORK APPLICATIONS TO RADAR DIRECTION FINDING# BEAMSTEERING# AND CONTROL. THIS APPROACH BAS BROAD APPLICATION TO DEFENSE. WE ANTICIPATE IMMEDIATE APPLICATION OF ADAPTIVE CONTROL MITH ONE SHOT LEARNING AND PERFECT MEMORY TO ROBOTICS WHERE EXPANSION AND DEFORHIATION OF ARM SEGMENTS IS A MAJOR CONCERN. ALSO# IGE INTEND TO APPLY THESE TECBNIQUES TO SPACE APPLICATIONS THAT INVOLVE DEFORIAATION OF STRESSED LINKS.