Research map / Large language models / Alignment and preferences
FGD-Align: Pluralistic Alignment for Large Language Models via Fuzzy Group Decision-Making
Weihang Pan, Zhengxu Yu, Yong Wu, Xun Liang, Zhongming Jin, Qiang Fu, Penghui Shang, Binbin Lin, Xiaofei He, Jieping Ye
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add triangular fuzzy numbers instead of a single scalar for annotator preferences (ablation) | FGD-Align · Multipref | Pref.Acc: 0.8550 -> 0.9000 | Improved |
| add triangular fuzzy numbers instead of a single scalar for annotator preferences (ablation) | FGD-Align · Multipref | AUROC: 0.8040 -> 0.8119 | Improved |
| add triangular fuzzy numbers instead of a single scalar for annotator preferences (ablation) | FGD-Align · Multipref | Div.PDS: 0.9189 -> 0.9214 | Improved |
| use adaptive fuzzy consensus strength instead of fixed label smoothing (cDPO, mu 0.1) (ablation) | PF-DPO · Multipref | Pref.Acc: 0.7400 -> 0.9000 | Improved |
| use adaptive fuzzy consensus strength instead of fixed label smoothing (cDPO, mu 0.2) (ablation) | PF-DPO · Multipref | Pref.Acc: 0.7100 -> 0.9000 | Improved |
| use fuzzy aggregation instead of Majority Vote data for DPO training | DPO · Multipref-LLMs | Pref.Acc: 0.7850 -> 0.8550 | Improved |
| use fuzzy aggregation instead of Copeland data for DPO training | DPO · Multipref-LLMs | AUROC: 0.5641 -> 0.6004 | Improved |
| use PF-DPO with fuzzy aggregation instead of standard DPO | Qwen2.5-Instruct-7B · Multipref | Pref.Acc: 0.7950 -> 0.9000 | Improved |
| use SimPO instead of DPO | Qwen2.5-Instruct-7B · Multipref | Div.PDS: 0.8998 -> 0.5974 | Hurt |
| use SimPO instead of DPO | Qwen2.5-Instruct-7B · Multipref | Pref.Acc: 0.7950 -> 0.8950 | Improved |
| scale model size from 1.5B to 32B | Qwen2.5-Instruct · Multipref | Pref.Acc vs detection accuracy | Mixed |
| train with FGD-Align instead of SimPO for diversity on disagreement prompts | Qwen2.5-Instruct-7B · Multipref disagreement prompts | Entropy: 1.3275 -> 6.8743 | 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 4 of 4 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
- On-the-fly Preference Alignment via Principle-Guided Decoding (ICLR 2025 · similarity 0.90)
- Linear Alignment: A Closed-form Solution for Aligning Human Preferences without Tuning and Feedback (ICML 2024 · similarity 0.90)
- Understanding the Learning Dynamics of Alignment with Human Feedback (ICML 2024 · similarity 0.90)
- Preference Ranking Optimization for Human Alignment (AAAI 2024 · similarity 0.90)
- Bounded Rationality for LLMs: Satisficing Alignment at Inference-Time (ICML 2025 · similarity 0.90)
- Panacea: Pareto Alignment via Preference Adaptation for LLMs (NeurIPS 2024 · similarity 0.90)
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.
