Fine-grained Approaches for Confidence Calibration of LLMs in Automated Code Revision
Published in arXiv preprint (under submission), 2026
Recommended citation: Hong Yi Lin, Chunhua Liu, Haoyu Gao, Patanamon Thongtanunam, and Christoph Treude. 2026. Fine-grained Approaches for Confidence Calibration of LLMs in Automated Code Revision. arXiv:2604.06723. https://arxiv.org/abs/2604.06723
Proposes local Platt-scaling over fine-grained confidence scores for automated code revision tasks (program repair, vulnerability repair, code refinement), consistently reducing calibration error across 3 tasks and 14 models.