AI Paper Atlas

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

ICML 2026 · Reinforcement learning › RL from feedback · approach: GRPO stability and variants

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 changedWhereMeasuredEffect
add feasibility moment constraint with anchor injection (ablation)CoPro · AnalogGym PFCHV: 0.550 -> 0.683Improved
add feasibility moment constraint with anchor injection (ablation)CoPro · AnalogGym PFCviolation rate: 0.260 -> 0.058Improved
use nonnegative target distribution instead of signed weights (ablation)CoPro · AnalogGym PFCGCI: 0.450 -> 0.021Improved
use nonnegative target distribution instead of signed weights (ablation)CoPro · AnalogGym PFCHV: 0.520 -> 0.683Improved
add distinctiveness moment constraint (ablation)CoPro · AnalogGym PFCGCI: 0.100 -> 0.021Improved
add distinctiveness moment constraint (ablation)CoPro · AnalogGym PFCviolation rate: 0.140 -> 0.058Improved
add dense fallback term to PF-progress score (ablation)CoPro · AnalogGym PFCHV: 0.570 -> 0.683Improved
add dense fallback term to PF-progress score (ablation)CoPro · AnalogGym PFCviolation rate: 0.170 -> 0.058Improved
replace GRPO signed advantages with CoPro projectionCoPro · AnalogGym PFCFOML: 221.7 -> 296.3Improved
replace GRPO signed advantages with CoPro projectionCoPro · AnalogGym DFCFC2FOML: 656.9 -> 976.3Improved
replace GRPO with CoProCoPro · BFCL-v3 (short length, Qwen-2.5-Instruct-1.5B)average accuracy: 28.3% -> 31.7%Improved
replace GRPO with CoProCoPro · 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.

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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy