This award is for the purchase of 64 TMS320C40 chips in order to build an asynchronous six-dimensional hypercube- based system to serve as a high-speed low-cost simulator of large neural networks. The system will be used to investigate neural network architectures and learning laws in applications where a great number of interconnection weights is necessary. Such applications include computer network congestion management, flow control and adaptive routing, and pattern recognition. New Jersey Institute of Technology is proposing a high-speed, low-cost solution to the tasks of designing and evaluating neural network architectures and learning protocols. They will be building an asynchronous six-dimensional hypercube- based system that supports efficiently both local and global communications, has a modest number of high-power processors, but is of low overall system cost.//
