Research map / Reinforcement learning / RL from feedback
Learning from Comparison: Constrained Projection Policy Optimization for Pareto-Front Improvement
Jintao Li, Maowen Tang, Yongji Long, Weixuan Liu, Yanlang Zheng, Sicheng He, Ao-Jin Li, Shui Yu, Yun 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 |
|---|---|---|---|
| add feasibility moment constraint with anchor injection (ablation) | CoPro · AnalogGym PFC | HV: 0.550 -> 0.683 | Improved |
| add feasibility moment constraint with anchor injection (ablation) | CoPro · AnalogGym PFC | violation rate: 0.260 -> 0.058 | Improved |
| use nonnegative target distribution instead of signed weights (ablation) | CoPro · AnalogGym PFC | GCI: 0.450 -> 0.021 | Improved |
| use nonnegative target distribution instead of signed weights (ablation) | CoPro · AnalogGym PFC | HV: 0.520 -> 0.683 | Improved |
| add distinctiveness moment constraint (ablation) | CoPro · AnalogGym PFC | GCI: 0.100 -> 0.021 | Improved |
| add distinctiveness moment constraint (ablation) | CoPro · AnalogGym PFC | violation rate: 0.140 -> 0.058 | Improved |
| add dense fallback term to PF-progress score (ablation) | CoPro · AnalogGym PFC | HV: 0.570 -> 0.683 | Improved |
| add dense fallback term to PF-progress score (ablation) | CoPro · AnalogGym PFC | violation rate: 0.170 -> 0.058 | Improved |
| replace GRPO signed advantages with CoPro projection | CoPro · AnalogGym PFC | FOML: 221.7 -> 296.3 | Improved |
| replace GRPO signed advantages with CoPro projection | CoPro · AnalogGym DFCFC2 | FOML: 656.9 -> 976.3 | Improved |
| replace GRPO with CoPro | CoPro · BFCL-v3 (short length, Qwen-2.5-Instruct-1.5B) | average accuracy: 28.3% -> 31.7% | Improved |
| replace GRPO with CoPro | CoPro · BFCL-v3 (long length, Qwen-2.5-Instruct-3B) | average accuracy: 38.1% -> 42.6% | 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.
- GDPO — GDPO: Group reward-Decoupled Normalization Policy Optimization for Multi-reward RL Optimization (ICML 2026) · 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
- Preference Conditioned Multi-Objective Reinforcement Learning: Decomposed, Diversity-Driven Policy Optimization (NeurIPS 2026 · similarity 0.89)
- Projection Optimization: A General Framework for Multi-Objective and Multi-Group RLHF (ICML 2025 · similarity 0.88)
- Preference Optimization on Pareto Sets: On a Theory of Multi-Objective Optimization (NeurIPS 2025 · similarity 0.88)
- An Offline Adaptation Framework for Constrained Multi-Objective Reinforcement Learning (NeurIPS 2024 · similarity 0.85)
- ACPO: A Policy Optimization Algorithm for Average MDPs with Constraints (ICML 2024 · similarity 0.85)
- On Generalization Across Environments In Multi-Objective Reinforcement Learning (ICLR 2025 · similarity 0.85)
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.
