Learning Based Patch Overfitting Detection: A Survey

Abstract

As the complexity of software increases, the requirements for software quality and security continue to increase. However, automatically generated fix patches may suffer from patch overfitting issues, leading to instability and security vulnerabilities. This paper aims to summarize the application of machine learning in detecting overfitting problems in automatic program repair (APR) patches. We review the current state of research on automatic program repair and overfitting detection, and we summarize the machine learning techniques employed. We identified a comprehensive list of available datasets and metrics commonly used in this research. Finally, this paper discusses the challenges faced in this field and systematically summarizes the application of machine learning in detecting patch overfitting problems in automatic program repair, providing a useful reference for future research.