Research map / Large language models / Alignment and preferences
Semi-Supervised Preference Optimization with Limited Feedback
Seonggyun Lee, Sungjun Lim, Seojin Park, Soeun Cheon, Kyungwoo Song
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 pseudo-labeled unpaired UltraChat data (SSPO) vs best baseline, 1% paired | Mistral-7B · AlpacaEval 2.0 LC (UltraFeedback) 1% | LC win rate: 18.2 -> 26.7 | Improved |
| SSPO with 1% paired vs baselines with 10% paired | Mistral-7B · AlpacaEval 2.0 LC (UltraFeedback) | LC win rate: 19.1 -> 26.7 | Improved |
| SSPO vs best baseline at 10% paired | Llama3-8B · AlpacaEval 2.0 LC (UltraFeedback) 10% | LC win rate: 16.7 -> 20.7 | Improved |
| SSPO vs best baseline at 1% paired | Llama3-8B · AlpacaEval 2.0 LC (UltraFeedback) 1% | LC win rate: 14.9 -> 14.8 | No clear effect |
| SSPO on MT-Bench vs baselines | Mistral/Llama3/Phi-2 · MT-Bench | score: 7.8 -> 7.7 | No clear effect |
| use adaptive scheduler instead of fixed gamma'=0.1 (ablation) | SSPO Mistral-7B 10% · AlpacaEval 2.0 LC (UltraFeedback) | LC win rate: 27.5 -> 30.0 | Improved |
| use adaptive scheduler instead of fixed gamma'=0.5 (ablation) | SSPO Mistral-7B 10% · AlpacaEval 2.0 LC (UltraFeedback) | LC win rate: 29.3 -> 30.0 | Improved |
| use prior 0.5 instead of 0.1 (ablation) | SSPO Mistral-7B 10% · AlpacaEval 2.0 LC (UltraFeedback) | LC win rate: 25.6 -> 30.0 | Improved |
| use prior 0.5 instead of 0.9 (ablation) | SSPO Mistral-7B 10% · AlpacaEval 2.0 LC (UltraFeedback) | LC win rate: 25.7 -> 30.0 | Improved |
| use SSPO instead of DPO on toy task | GPT-2-small toy, n_L=10 · toy dataset noise 0% | test accuracy: 0.743 -> 0.841 | Improved |
| use SSPO instead of DPO with 50% label noise and n_L=100 | GPT-2-small toy · toy dataset noise 50% | test accuracy: 0.554 -> 0.563 | No clear effect |
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
- RePO: Understanding Preference Learning Through ReLU-Based Optimization (NeurIPS 2025 · similarity 0.91)
- Proximalized Preference Optimization for Diverse Feedback Types: A Decomposed Perspective on DPO (NeurIPS 2025 · similarity 0.90)
- Sequential Preference Optimization: Multi-Dimensional Preference Alignment with Implicit Reward Modeling (AAAI 2025 · similarity 0.90)
- Less is More: Improving LLM Alignment via Preference Data Selection (NeurIPS 2025 · similarity 0.90)
- Data with High and Consistent Preference Difference Are Better for Reward Model (AAAI 2025 · similarity 0.90)
- Reward-Augmented Data Enhances Direct Preference Alignment of LLMs (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.
