Research map / Large language models / Alignment and preferences
Aligning Language Models with Human Preferences via a Bayesian Approach
Jiashuo WANG, Haozhao Wang, Shichao Sun, Wenjie Li
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 |
|---|---|---|---|
| use d-PM instead of majority-vote preference model (ablation) | Aligned MultiESC · ESConv | BLEU-4: 2.88 -> 3.36 | Improved |
| use d-PM instead of soft-label preference model (ablation) | Aligned MultiESC · ESConv | CIDEr: 30.75 -> 32.65 | Improved |
| use d-PM instead of unaggregated annotations (w/oA) (ablation) | Aligned MultiESC · ESConv | CIDEr: 28.58 -> 32.65 | Improved |
| align MultiESC with majority-vote preference model (ablation) | MultiESC · ESConv | BLEU-4: 3.14 -> 2.88 | Hurt |
| align MultiESC with d-PM calibration | MultiESC · ESConv human eval | Global Consensus: 2.693 -> 2.783 | Improved |
| align Blender-Vanilla with d-PM | Blender-Vanilla · ESConv | Extreme: 51.02 -> 50.38 | Hurt |
| use majority-vote model for Blender-Joint (ablation) | Blender-Joint · ESConv | CIDEr: 16.91 -> 21.86 | Improved |
| use w/oA model for Blender-Joint (ablation) | Blender-Joint · ESConv | CIDEr: 16.91 -> 15.29 | Hurt |
| align T5-small with d-PM (greedy) | T5-small · MIC | ROUGE-1: 37.04 -> 38.15 | Improved |
| align BART-large with soft model (beam) (ablation) | BART-large · MIC | ROUGE-2: 35.07 -> 34.85 | Hurt |
| align Flan-T5 with d-PM (beam) | Flan-T5-base · MIC | ROUGE-1: 55.07 -> 55.18 | Improved |
| align BART-large with d-PM (beam) | BART-large · MIC | Avg.Len: 9.44 -> 9.40 | No clear effect |
Closest papers by meaning
- Aligning Language Models Using Follow-up Likelihood as Reward Signal (AAAI 2025 · similarity 0.91)
- Preference Ranking Optimization for Human Alignment (AAAI 2024 · similarity 0.91)
- Sequential Preference Optimization: Multi-Dimensional Preference Alignment with Implicit Reward Modeling (AAAI 2025 · similarity 0.90)
- Transforming and Combining Rewards for Aligning Large Language Models (ICML 2024 · similarity 0.90)
- Beyond Bradley-Terry Models: A General Preference Model for Language Model Alignment (ICML 2025 · similarity 0.90)
- Latent Preference Coding: Aligning Large Language Models via Discrete Latent Codes (ICML 2025 · 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.
