Dynamic competitive learning applied to the clone detection problem

Abstract

A novel neural network architecture, 'Dynamic Competitive Learning', has been applied to the software clone detection problem in telecommunications systems. A software clone is a copy (subsequently modified), of a piece of existing code; it is common practice in large systems to obtain similar code via reuse of fully tested code. This paper discusses a hierarchical network with dramatically reduced training time compared to a Self Organising Map, but with at least as good a classification performance.