Research map / Reinforcement learning / RL from feedback
Swap-guided Preference Learning for Personalized Reinforcement Learning from Human Feedback
Gihoon Kim, Euntai Kim
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 |
|---|---|---|---|
| replace VPL with SPL | VPL · UF-P-4 (Llama-3.2-3B) | accuracy (%): 57.03 -> 61.56 | Improved |
| replace VPL with SPL | VPL · UF-P-4 (Llama-3.1-8B) | accuracy (%): 57.14 -> 62.21 | Improved |
| replace VPL with SPL | VPL · UF-P-2 (Llama-3.1-8B) | accuracy (%): 62.66 -> 63.71 | Improved |
| replace VPL with SPL | VPL · Pets (Llama-3.1-8B) | accuracy (%): 75.33 -> 100.0 | Improved |
| replace VPL with SPL | VPL · Pets (Llama-3.2-3B) | accuracy (%): 99.67 -> 100.0 | Improved |
| vary KL weight beta from 3e-7 to 3e-5 with SPL instead of VPL | VPL · UF-P-4 active units | active units (%): 0.00 -> 77.15 | Improved |
| add P-IAF on top of L_guide (ablation) | SPL · UF-P-4 (Llama-3.1-8B) | accuracy (%): 57.18 -> 62.14 | Improved |
| add swap-guided L_guide to P-IAF (ablation) | SPL · UF-P-4 (Llama-3.1-8B) | accuracy (%): 59.10 -> 62.14 | Improved |
| add adaptive latent conditioning to L_guide + P-IAF (ablation) | SPL · UF-P-4 (Llama-3.1-8B) | accuracy (%): 62.14 -> 62.21 | Improved |
| replace standard IAF with P-IAF (ablation) | SPL-IAF · UF-P-4 (Llama-3.1-8B) | accuracy (%): 60.56 -> 62.21 | Improved |
| replace standard IAF with P-IAF (ablation) | SPL-IAF · UF-P-4 (Llama-3.2-3B) | accuracy (%): 59.50 -> 61.56 | Improved |
| add standard IAF to VPL (ablation) | VPL · UF-P-4 (Llama-3.2-3B) | accuracy (%): 57.03 -> 58.01 | Improved |
Closest papers by meaning
- Personalizing Reinforcement Learning from Human Feedback with Variational Preference Learning (NeurIPS 2024 · similarity 0.93)
- Iterative Preference Learning from Human Feedback: Bridging Theory and Practice for RLHF under KL-constraint (ICML 2024 · similarity 0.89)
- Data with High and Consistent Preference Difference Are Better for Reward Model (AAAI 2025 · similarity 0.89)
- Online Iterative Reinforcement Learning from Human Feedback with General Preference Model (NeurIPS 2024 · similarity 0.88)
- MaxMin-RLHF: Alignment with Diverse Human Preferences (ICML 2024 · similarity 0.88)
- MPO: An Efficient Post-Processing Framework for Mixing Diverse Preference Alignment (ICML 2025 · similarity 0.88)
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.
