A parallel stochastic algorithm for learning logic expressions under noise

Abstract

The problem of learning conjunctive and disjunctive concepts from a series of positive and negative examples of the concept is considered. Formulating the problem In the Probably Approximalely Correct(PAC) Learning framework . the goal of such inductive learning is precisely characteried . A parallel distributed stochastic algorithm is presented . It has been proved that the algorithm learns the class of simple conjunctive concepts in the presence of Upto 50% unbiased noise over both nominal and linear attributes. As an extension to this an algorithm that learns a class of disjunctive concepts is proposed. Through empirical studies it is seen that the algorithm is quite efficient for learning conjunctive concepts and certain disjunctive concepts.