Metraipolis › archive › dc_80666c28b4ef
CalibraRL: Confidence-Calibrated Reinforcement Learning for Mitigating Hallucinations in RAG — arXiv:2610.35142
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.35142
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-27T08:12:12.962Z
- SHA-256 of the archived text
efe849e77913cf6a8a29683b5737d14dca0edb0affb3f957ec6bc0ad744e2d9e- Stable URL of this page
https://metraipolis.com/documento/dc_80666c28b4ef
How to cite
CalibraRL: Confidence-Calibrated Reinforcement Learning for Mitigating Hallucinations in RAG — arXiv:2610.35142 — arXiv (en), https://arxiv.org/abs/2610.35142 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_80666c28b4ef (Metraipolis archive dc_80666c28b4ef, sha256 efe849e77913cf6a8a29683b5737d14dca0edb0affb3f957ec6bc0ad744e2d9e)
BibTeX
@misc{dc80666c28b4ef,
title = {CalibraRL: Confidence-Calibrated Reinforcement Learning for Mitigating Hallucinations in RAG — arXiv:2610.35142},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.35142},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_80666c28b4ef; sha256 efe849e77913cf6a8a29683b5737d14dca0edb0affb3f957ec6bc0ad744e2d9e},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.35142. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Large language models (LLMs) often produce plausible but incorrect answers, motivating reinforcement learning methods that reward correct answers, penalize incorrect answers, and allow abstention. However, fixed outcome-based rewards do not distinguish incorrect responses with different confidence levels or account for differences in answering reliability across queries. Consequently, such rewards may inadequately penalize overconfident errors and fail to discourage unnecessary abstention. We propose CalibraRL, a reinforcement learning framework that incorporates response confidence and query-level answering evidence into reward design. For non-abstaining responses, CalibraRL combines correctness-based utility with a Brier-based penalty that encourages agreement between confidence and correctness. For abstentions, it estimates conditional answering reliability using a Beta posterior derived from the correctness of non-abstaining responses within each sampled group. The abstention reward increases with the posterior probability that the model's accuracy on non-abstaining responses is below a predefined threshold. CalibraRL reuses training rollouts without requiring additional annotations indicating whether the model knows the answer to each question. Experiments on three QA datasets demonstrate that CalibraRL improves answer precision and reduces hallucinations compared with strong baselines. Further analysis shows that CalibraRL reduces mean expected calibration error (ECE) by 30.8\% across two backbone models and three datasets, highlighting its potential to improve the reliability of retrieval-augmented generation (RAG) in web search. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno). Categorie arXiv: Computation and Language (cs.CL). Pagina del lavoro: https://arxiv.org/abs/2610.35142 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-27T07:56:00.254Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).