HNETTER-a heuristically driven neural reasoning system
Kaplan, Johnson · 1988
A description is given of a probabilistic reasoning system that attains understanding of simple 2D shapes and their properties from a small number of possibly incorrect examples. The system represents a fusion of heuristic reasoning and artificial neural nets with the former accomplishing generation and pruning of hypotheses and the latter assimilating the arbitrarily complex combinations of relevant hypotheses. The authors discuss the relative advantages of both approaches and suggest how, based on their experience, a learning task can be successfully split between them.>