Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Learn about PRISM, a novel technique for detecting overfitting patches in automatic program repair (APR) systems, presented as a 14-minute conference talk from OOPSLA 2025. Discover how researchers Dowon Song and Hakjoo Oh from Korea University address the persistent problem of overfitting patches—those that pass test suites but fail to actually fix bugs—which degrade APR performance and increase developer review burden. Explore the innovative semantic features that capture patch-induced behavioral changes and understand the tailored learning algorithm designed to preserve correct patches while filtering out incorrect ones. Examine experimental results from 10 APR tools demonstrating how PRISM uniquely reduces review burden while finding more correct patches, unlike other methods that lower fix rates by misclassifying correct patches. Analyze the evaluation findings from 1,829 labeled patches showing PRISM's superior ability to remove incorrect patches while maintaining equal correct patch preservation rates. Access supplementary materials including the research article, artifact archive, and ORCID profiles for comprehensive understanding of this semantic-based approach to improving automatic program repair systems.
