AI Paper Atlas

Research map / Large language models / Alignment and preferences

Mind the Gap: Structure-Aware Consistency in Preference Learning

Mehryar Mohri, Yutao Zhong

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

Read the paper

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
use structure-aware margin (SA-DPO) instead of fixed-margin SimPODPO · UltraFeedback hard ambiguous subsetranking accuracy: 0.650 -> 0.700Improved
use structure-aware margin (SA-DPO) instead of plain DPODPO · UltraFeedback hard ambiguous subsetranking accuracy: 0.590 -> 0.700Improved
use structure-aware margin (SA-DPO) instead of DPODPO · UltraFeedback distinct splitranking accuracy: 0.766 -> 0.790Improved
use structure-aware margin (SA-DPO) instead of DPODPO · Argilla DPO-Mix-7k Qwen2.5-7Branking accuracy (distinct / ambiguous): 0.774 -> 0.798 distinct; 0.727 -> 0.746 ambiguousImproved
add semantic-distance-scaled margin (SA-DPO) on near-synonym pairs (ablation)DPO · Synonym stress testfinal loss: 0.1695 -> 0.003Improved
replace logistic loss with squared (IPO) loss under capacity-limited LoRA (ablation)DPO · HH-RLHF, gamma=1.0final ranking accuracy: 71.6% -> 94.5%Improved
replace logistic loss with cubic hinge loss under capacity-limited LoRA (ablation)DPO · HH-RLHF, gamma=1.0final ranking accuracy: 71.6% -> 99.7%Improved
add structure-aware margin (SA-DPO) versus DPO for margin satisfactionDPO · Qwen2.5-7Bmargin satisfaction rate: 58.0% -> 72.3%Improved
use fixed margin (SimPO) instead of DPO for margin satisfactionDPO · Qwen2.5-7Bmargin satisfaction rate: 58.0% -> 63.0%Improved
SA-DPO versus standard DPO in head-to-head generationDPO · MT-Bench-style prompts, PairRMwin-rate: 58.5% win-rateImproved

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