PARALLEL NEURAL NETWORK TOOLBOX FOR SUPERCOMPUTERS

THIS PROPOSAL INVOLVES RESEARCH INTO A SUPERCOMPUTER-BASED ENVIRONMENT FOR RESEARCH INTO MASSIVELY PARALLEL NEURAL-NETWORK ALGORITHMS FOR LARGE SCALE, COMPUTATIONALLY INTENSE PROBLEMS. THIS ENVIRONMENT WILL HAVE APPLICATION TO AREAS SUCH AS IMAGE PROCESSING, PATTERN RECOGNITION, CONTROL SIGNAL PROCESSING, DATA COMPRESSION, AND FAULT DIAGNOSIS. ALL OF THESE ARE AREAS WHICH INVOLVE MASSIVE AMOUNTS OF DATA AND REQUIRE EXTREMELY THROUGHPUT SPEEDS. WE WILL INVESTIGATE METHODS OF PARALLELIZING EXISTING ALGORITHMS IN OUR NEURAL NETWORK TOOLBOX FO RUSE ON TH ESILICON GRAPHICS LINE OF WORKSTATIONS, THE TI-TMS320C40, AND THE CRAY 2. WE WILL ALSO INVESTIGATE THE APPLICATION OF A SILICOAN GRAPHICS BASED EXPLORER ENVIRONMENT OF THE CRAY. THIS ENVIRONMENT WILL ALLOW THE MANIPULATION AND TESTING OF BOTH SINGLE NEURAL NETWORK ALGORITHMS AND YSSTEM SOF SUCH ALGORITHMS. IT WILL ALSO PROVIDE A CONSISTENT AND USER FRIENDLY ENVIRONMENT FOR RESEARCH.