Explainable Automated Program Repair

Automated Program Repair aims to assist developers by automatically generating patches for incorrect code, thereby saving significant debugging time. While this can be helpful, the missing part is that due to the black-box nature of some such tools, the developer cannot know why the model generated that patch. This issue can be addressed by applying an explainable AI technique. However, existing techniques do not take the nature of source code into account. We address this in our research by considering the syntax and semantics of a programming language, generating perturbations based on that, and using those perturbations to identify the parts of the incorrect code that contributed the most to the model's output.