Scalable and accurate detection of semantic clones

Abstract

Code clone detection plays an important role in software maintenance and evolution. There are many new applications emerging that rely on clones detected across software systems, and hence to address this, many code clone detection tools are being developed. However only a few of them target semantic clones. With deep learning taking a new turn today, extensive work has started to leverage these models to detect clones. These models use lexical information and syntactic structures like the abstract syntax trees to detect the clones, however, these methods do not take into account the available structural and semantic information that the codes offer and this limits the capabilities of such methods. Using Program Dependence Graphs and attention based learning, we want to fully leverage the structured syntactic and semantic information and develop a tool which can be used to detect the clone which might differ syntactically but yield the same semantics.