This research studies the theory and implementation of knowledge representation and reasoning on a massively parallel "connectionist" computer system. The focus is on specific tasks: inheritance of properties by objects in a subclass from properties of the superclass, pattern recognition, and pattern completion. The goal is to employ the advantages of massively parallel systems - tolerating noise and ambiguity and operating rapidly, e.g. for real-time applications. The importance of this research is that massively parallel systems - employing tens of thousands of computing elements - may provide significant new computing power to artificial intelligence and other applications. However, detailed theory and practical design methods for these systems must first be developed.
