This research revolves around the acquisition of a Single Instruction, Multiple Data (SIMD) computer and its applications to a variety of problems. The problems include: parallel group algorithms, natural language processing, neural network learning algorithms, and the scientist's assistant project. The SIMD computer is uniquely suited as both an exploratory platform upon which to test new algorithms and as a compute server to actually solve problems that would otherwise require substantial super computer resources to solve. The next generation of super computers are likely to be highly parallel computers following the Single Instruction, Multiple Data (SIMD) architecture and/or the Multiple Instruction, Multiple Data (MIMD) architecture. In either case, algorithms need to be developed and tested for the architectures to see what problems run best on what architectures. This can be done in two ways. The first is to pick arbitrary problems and develop algorithms for both architectures and then compare the results. The second method is to look for problem classes that are best solved by a particular architecture. The group at Northeastern have chosen the second alternative and are looking for SIMD algorithms for problems that seem to be particularly well suited for that architecture. Two problems in particular are promising areas of research. The parallel group algorithms are intended to solve computationally intensive problems that could not easily be solved on a single processor architecture. The scientist's assistant project is a sophisticated very large pattern matching application that is particularly well suited to the SIMD architecture.
