Program Repair

Automated program repair (APR) aims to automatically generate code patches to fix software bugs, reducing development time and costs. Current research heavily utilizes large language models (LLMs), often employing techniques like fine-tuning, prompt engineering, and reinforcement learning to enhance patch generation accuracy and efficiency, sometimes incorporating feedback loops and iterative refinement strategies. The field is actively developing improved benchmarks and datasets to address data leakage concerns and better evaluate model performance, ultimately aiming to improve software reliability and security.