Guiding term reduction through a neural network: Some preliminary results for the group theory

Abstract

Some experiments have been carried out in order to build Neural Networks which, given a term belonging to an Equational Theory, could suggest which rewrite rules belonging to the Completed TRS for that theory represent the best choice at each reduction step in order t6 minimize the number of reductions needed to reach the normal form. For the Groups Theory a net was built which had an accuracy of 61%. Moreover the same net in the 71% of the cases could correctly suggest a rule applicable to the term.