LLMs for Code-Change-Related Tasks
Evaluating and improving large language models on code-change understanding tasks.
This research line studies whether large language models understand code changes, not only standalone code. It covers tasks such as code review generation, commit message generation, and just-in-time comment update, and examines how in-context learning and parameter-efficient tuning help LLMs adapt to change-centric software engineering scenarios.
The core motivation is that realistic development work is often about understanding differences between code versions. My work aims to make code LLMs more reliable in these change-aware workflows.