AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Anchored Policy Optimization: Mitigating Exploration Collapse via Support-Constrained Rectification

Tianyi Wang, Long Li, Hongcan Guo, Yibiao Chen, Yixia Li, Yong Wang, Yun Chen, Guanhua Chen

ICML 2026 · Reinforcement learning › RL from feedback · approach: Scaling RL with verifiable rewards · 0 citations (OpenAlex)

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
replace GRPO with APO (anchored ratio rectification)Qwen2.5-Math-7B · 5 math benchmarks avgPass@1: 48.68 -> 50.67Improved
replace GRPO with APOQwen2.5-Math-7B · 5 math benchmarks avgPass@K: 76.56 -> 78.07Improved
replace GRPO with APOLlama-3.2-3B-Instruct · 5 math benchmarks avgPass@K: 43.82 -> 47.11Improved
replace GRPO with APOQwen2.5-7B · 5 math benchmarks avgPass@1: 42.68 -> 43.57Improved
add global KL regularization to GRPO (ablation)GRPO on Qwen2.5-Math-7B · 5 math benchmarks avgPass@1: 48.68 -> 44.53Hurt
apply KL penalty only on errors instead of globally (ablation)GRPO-KL on Qwen2.5-Math-7B · 5 math benchmarks avgPass@1: 44.53 -> 46.79Improved
error-only KL versus unregularized GRPO (ablation)GRPO on Qwen2.5-Math-7B · 5 math benchmarks avgPass@1: 48.68 -> 46.79Hurt
add negative sample reinforcement (NSR) to GRPOQwen2.5-Math-7B · 5 math benchmarks avgPass@1: 48.68 -> 48.58No clear effect
remove pull anchor (beta = 0) (ablation)APO on Qwen2.5-Math-7B · 5 math benchmarks avgPass@1: 50.67 -> 48.86Hurt
reduce anchor set size to K=4 from 8 (ablation)APO on Qwen2.5-Math-7B · 5 math benchmarks avgPass@1: 50.67 -> 49.46Hurt
broaden anchor set to K=16 from 8 (ablation)APO on Qwen2.5-Math-7B · 5 math benchmarks avgPass@1Hurt
push coefficient lambda 1.0 instead of 1.05 (ablation)APO on Qwen2.5-Math-7B · 5 math benchmarks avgPass@1: 50.67 -> 49.24Hurt

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