Research map / Large language models / Alignment and preferences
Mind the Gap: Structure-Aware Consistency in Preference Learning
Mehryar Mohri, Yutao Zhong
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 changed | Where | Measured | Effect |
|---|---|---|---|
| use structure-aware margin (SA-DPO) instead of fixed-margin SimPO | DPO · UltraFeedback hard ambiguous subset | ranking accuracy: 0.650 -> 0.700 | Improved |
| use structure-aware margin (SA-DPO) instead of plain DPO | DPO · UltraFeedback hard ambiguous subset | ranking accuracy: 0.590 -> 0.700 | Improved |
| use structure-aware margin (SA-DPO) instead of DPO | DPO · UltraFeedback distinct split | ranking accuracy: 0.766 -> 0.790 | Improved |
| use structure-aware margin (SA-DPO) instead of DPO | DPO · Argilla DPO-Mix-7k Qwen2.5-7B | ranking accuracy (distinct / ambiguous): 0.774 -> 0.798 distinct; 0.727 -> 0.746 ambiguous | Improved |
| add semantic-distance-scaled margin (SA-DPO) on near-synonym pairs (ablation) | DPO · Synonym stress test | final loss: 0.1695 -> 0.003 | Improved |
| replace logistic loss with squared (IPO) loss under capacity-limited LoRA (ablation) | DPO · HH-RLHF, gamma=1.0 | final ranking accuracy: 71.6% -> 94.5% | Improved |
| replace logistic loss with cubic hinge loss under capacity-limited LoRA (ablation) | DPO · HH-RLHF, gamma=1.0 | final ranking accuracy: 71.6% -> 99.7% | Improved |
| add structure-aware margin (SA-DPO) versus DPO for margin satisfaction | DPO · Qwen2.5-7B | margin satisfaction rate: 58.0% -> 72.3% | Improved |
| use fixed margin (SimPO) instead of DPO for margin satisfaction | DPO · Qwen2.5-7B | margin satisfaction rate: 58.0% -> 63.0% | Improved |
| SA-DPO versus standard DPO in head-to-head generation | DPO · MT-Bench-style prompts, PairRM | win-rate: 58.5% win-rate | Improved |
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.
- SimPO — SimPO: Simple Preference Optimization with a Reference-Free Reward (NeurIPS 2024) · better on 6 of 6 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
- Less is More: Improving LLM Alignment via Preference Data Selection (NeurIPS 2025 · similarity 0.90)
- Towards Robust Alignment of Language Models: Distributionally Robustifying Direct Preference Optimization (ICLR 2025 · similarity 0.89)
- 3D-Properties: Identifying Challenges in DPO and Charting a Path Forward (ICLR 2025 · similarity 0.89)
- Learning Where It Matters: Geometric Anchoring for Robust Preference Alignment (NeurIPS 2026 · similarity 0.89)
- Reward-Augmented Data Enhances Direct Preference Alignment of LLMs (ICML 2025 · similarity 0.89)
- Noise Contrastive Alignment of Language Models with Explicit Rewards (NeurIPS 2024 · similarity 0.89)
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.
