AI Paper Atlas

Research map / Large language models / Alignment and preferences

Unbiased Alignment for Large Language Models with Noisy Preferences

Jialiang Wang, Xianming Liu, Xiong Zhou, Hui Liu, Haoliang Li

ICML 2026 · Large language models › Alignment and preferences · approach: Robust and calibrated DPO · 0 citations (OpenAlex)

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
replace DPO loss with UDPO lossDPO, Llama-3.2-3B · HH+TL;DR averagewin rate vs SFT (%): 71.4 -> 85.9Improved
replace DPO loss with UDPO lossDPO, Qwen-3-1.7B · HH+TL;DR averagewin rate vs SFT (%): 65.9 -> 79.4Improved
replace RM loss with URM lossRM, Qwen-3-1.7B · HH+TL;DR averagewin rate vs SFT (%): 63.9 -> 70.1Improved
replace RM loss with URM lossRM, Llama-3.2-3B · HH+TL;DR averagewin rate vs SFT (%): 75.5 -> 79.4Improved
replace RM loss with URM lossRM, Qwen-3-8B · UltraFeedbackwin rate vs SFT (%): 54.3 -> 64.6Improved
replace DPO loss with UDPO lossDPO, Llama-3.1-8B · UltraFeedbackwin rate vs SFT (%): 65.4 -> 74.3Improved
replace DPO loss with UDPO lossDPO, Qwen-3-8B · UltraFeedbackwin rate vs SFT (%): 56.3 -> 60.6Improved
add label smoothing (cRM, cDPO) to the lossRM and DPO, Llama-3.2-3B · HH+TL;DR averagewin rate vs SFT (%): 75.5 -> 75.7 (RM); 71.4 -> 70.7 (DPO)No clear effect
add label smoothing to DPO (cDPO)DPO, Qwen-3-1.7B · HH+TL;DR averagewin rate vs SFT (%): 65.9 -> 63.6Hurt
use URM instead of rRM under 40% flipped labelsrRM, Llama-3.2-3B · TL;DR 40% flipwin rate vs SFT (%): 68.6 -> 77.5Improved
use UDPO instead of rDPO under instance-dependent noise 40%rDPO, Qwen-3-1.7B · HH IDN 40%win rate vs SFT (%): 63.7 -> 68.0Improved
use URM instead of RM under instance-dependent noise 40%RM, Qwen-3-1.7B · HH IDN 40%win rate vs SFT (%): 46.0 -> 54.6Improved

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