Research map / Large language models / Alignment and preferences
Reward-free Alignment for Conflicting Objectives
Peter Chen, Xiaopeng Li, Xi Chen, Tianyi Lin
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 clipping to the CAGrad correction weights (ablation) | CAGrad · Reddit summary quality-conciseness (validation margins) | preference margin | Improved |
| add clipping to the CAGrad correction weights at extreme weights (0.8 or 0.2) (ablation) | CAGrad · Reddit summary Pareto frontier | Pareto trade-off | Improved |
| add clipping to the CAGrad correction weights at balanced weight 0.5 (ablation) | CAGrad · Reddit summary Pareto frontier | Pareto trade-off | No clear effect |
| increase CAGrad radius c within 0.35, 0.4, 0.45 (ablation) | RACO · Reddit summary quality (Qwen3) | quality margin | Mixed |
| use conflict-aware RACO instead of weighted-sum DPO LW | DPO LW · BeaverTails Qwen3-4B-Instruct (w_help=0.2) | win rate vs RACO (%): 63.86 | Improved |
| use RACO instead of AMoPO | AMoPO · BeaverTails Qwen3-4B-Instruct (w_help=0.2) | win rate (%): 78.92 | Improved |
| use RACO instead of AMoPO | AMoPO · BeaverTails Qwen3-4B-Base (w_help=0.2) | win rate (%): 43.37 | Hurt |
| use RACO instead of AMoPO | AMoPO · BeaverTails Gemma3-4B-Instruct (w_help=0.2) | win rate (%): 45.78 | No clear effect |
| use RACO instead of DPO LW | DPO LW · BeaverTails Gemma3-4B-Base (w_help=0.8) | win rate (%): 70.48 | Improved |
| use RACO instead of AMoPO | AMoPO · BeaverTails Gemma3-4B-Base (w_help=0.8) | win rate (%): 76.51 | Improved |
Closest papers by meaning
- MATO: Multi-objective Personalized Alignment with Test-time Optimization for Large Language Models (NeurIPS 2026 · similarity 0.93)
- MetaAligner: Towards Generalizable Multi-Objective Alignment of Language Models (NeurIPS 2024 · similarity 0.93)
- TODO: Enhancing LLM Alignment with Ternary Preferences (ICLR 2025 · similarity 0.92)
- Panacea: Pareto Alignment via Preference Adaptation for LLMs (NeurIPS 2024 · similarity 0.92)
- ComPO: Preference Alignment via Comparison Oracles (NeurIPS 2025 · similarity 0.91)
- Alignment of Large Language Models with Constrained Learning (NeurIPS 2025 · similarity 0.91)
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.
