AI Paper Atlas

Research map / Large language models / Alignment and preferences

Towards Disentangled Preference Optimization Dynamics: Suppress the Loser, Preserve the Winner

Wei Chen, Yubing Wu, Junmei Yang, Delu zeng, Qibin Zhao, John Paisley, Min Chen, Zhou Wang

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
add reward calibration (RC)DPO · UltraFeedback, Mistral-7Baverage score: 24.26 -> 36.44Improved
add reward calibration (RC)DDRO · UltraFeedback, Mistral-7Baverage score: 16.53 -> 36.34Improved
add reward calibration (RC)SimPO, BCE, LSIF, TI-DPO (strong baselines) · UltraFeedback, Mistral-7Baverage score: SimPO +0.01, BCE -0.05, LSIF +0.15, TI-DPO +0.00No clear effect
add reward calibration (RC)DPO, SimPO, BCE, LSIF, TI-DPO · UltraFeedback, Qwen2.5-7Baverage score: DPO +0.48, SimPO +0.26, BCE +1.33, LSIF +0.23, TI-DPO +0.11Improved
add reward calibration (RC)DDRO · UltraFeedback, Qwen2.5-7Baverage score: 45.25 -> 45.20Hurt
add reward calibration (RC)DPO · UltraFeedback, Pythia-2.8BARC: 34.04 -> 37.12Improved
add reward calibration (RC)SimPO · UltraFeedback, Pythia-2.8BARC: 32.59 -> 37.03Improved
add reward calibration (RC)CPO · UltraFeedback, Pythia-2.8BMMLU-ProHurt
add reward calibration (RC)DPO · AlpacaEval 2.0, Pythia-2.8BLC win rate: 70.56%Improved
add reward calibration (RC)DPO, BCE · Anthropic-HH, Mistral-7Baverage score: DPO 36.04 -> 35.93; BCE 35.99 -> 36.09No clear effect
vary log-domain EMA momentum from 0.5 to 0.999 (ablation)RC with DPO · UltraFeedback, Pythia-2.8Baverage score: 26.91 (w/o RC) -> 28.43 to 28.95No clear effect
train for 2 epochs instead of 1 (ablation)RC with DPO · UltraFeedback, Mistral-7Baverage score: 36.84 -> 36.93No clear effect

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