Inductive Inference in Theo

Statistical ordering of problem-solving methods is accomplished by means of an inductive learning system called SE, which orders by means of a random sample of executed computations, as well as a predicate language for defining clusters of computations (classes of problems). The function used for ordering is:

Cost_of_success+ (probability_of_failure/probabilty_of_success)* cost_of_failure.

SE accepts as input a slot address and the corresponding list of AVAILABLE.METHODS for that slot, and produces a list of methods ordered in the sequence in which they should be attempted. SE makes use of Theo's meta-knowledge in order to make these method orderings. These methods have been proven to minimize the expected inference cost for Theo. The process, however, depends upon the reliablity of the predictions generated by SE.

Return to the top of this architecture.

Go to a discussion of this capability for multiple architectures.

Current Location: Theo-Capabilities-Inductive-Inference

Go to NEXT page.