Abstract
A way of acquiring fuzzy inference rules is introduced based on the novel notion of fuzziness referred to as constraint-interval fuzzy set. The basic apparatus of this rule acquisition are a method of generalizing instance data (instance information) and also a method to organize and to refine these generalized pieces of instance information. For the instance generalization, we will introduce a new method utilizing pairs of instance data, which we call DIG (Double Instance Generalization). This method is based on the notion of convexity of predicates. For the organization and refinements of generalized pieces of information, we will employ a Hopfield type neural network, by which effective ways of extracting fuzzy inference rules from a collection of instance data on fuzzy constraints or crisp constraints are derived.
