AI Paper Atlas

Research map / Large language models / Alignment and preferences

Robust Preference Optimization: Aligning Language Models with Noisy Preference Feedback

Xiaoyang Cao, Zelai Xu, Mo Guang, Kaiwen Long, Michiel A. Bakker, Yu Wang, Chao Yu

ICLR 2026 · Large language models › Alignment and preferences · approach: Active and token-level preference learning

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add RE-PO reliability reweighting to DPODPO on Mistral-7B-Instruct-v0.2 · AlpacaEval 2 (UltraFeedback-trained)LC / WR (%): 28.5/28.6 -> 35.5/33.0Improved
add RE-PO to IPOIPO on Llama-3-8B-Instruct · AlpacaEval 2 (UltraFeedback-trained)LC / WR (%): 43.6/41.6 -> 48.3/48.6Improved
add RE-PO to DPODPO on Llama-3-8B-Instruct · AlpacaEval 2 (UltraFeedback-trained)LC / WR (%): 40.8/42.9 -> 44.1/46.2Improved
add RE-PO to SimPOSimPO on Llama-3-8B-Instruct · AlpacaEval 2 (UltraFeedback-trained)LC / WR (%): 44.5/37.1 -> 46.9/39.4Mixed
add RE-PO to CPOCPO on Mistral-7B-Instruct-v0.2 · AlpacaEval 2 (UltraFeedback-trained)LC / WR (%): 26.3/26.4 -> 27.6/27.8No clear effect
add label smoothing to SimPOSimPO on Llama-3-8B-Instruct · AlpacaEval 2 (UltraFeedback-trained)LC (%): 44.5 -> 48.1Improved
add label smoothing to IPOIPO on Llama-3-8B-Instruct · AlpacaEval 2 (UltraFeedback-trained)LC / WR (%): 43.6/41.6 -> 40.3/38.2Hurt
use RE-DPO instead of rDPO and Holder-DPODPO on Llama-3-8B-Instruct · AlpacaEval 2 (UltraFeedback-trained)LC / WR (%): 37.3/35.4 -> 44.1/46.2Improved
add RE-DPO with per-annotator reliabilities on MultiPrefDPO on Llama-3-8B-Instruct · AlpacaEval 2 (MultiPref-trained)LC / WR (%): 36.7/39.3 -> 41.1/44.4Improved
add RE-DPO with per-annotator reliabilities on MultiPrefDPO on Mistral-7B-Instruct-v0.2 · AlpacaEval 2 (MultiPref-trained)LC / WR (%): 28.8/26.4 -> 31.8/28.8Improved
set initial reliability eta0 to 0.9 instead of 0.99 (ablation)RE-DPO Mistral-7B · AlpacaEval 2 (UltraFeedback-trained)LC (%): 30.9 -> 35.5Improved
set initial reliability eta0 to 0.9 instead of 0.55 (ablation)RE-DPO Mistral-7B · AlpacaEval 2 (UltraFeedback-trained)LC (%): 31.4 -> 35.5Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy