AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

On Group Relative Policy Optimization Collapse in Agent Search: The Lazy Likelihood-Displacement

Wenlong Deng, Yushu Li, Boying Gong, Yi Ren, Christos Thrampoulidis, Xiaoxiao Li

ICML 2026 · Reinforcement learning › RL from feedback · approach: GRPO stability and variants · 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
add LLDS likelihood-preserving regularizerGRPO, Qwen2.5-3B-Base · 7 QA benchmarks avgEM: 0.312 -> 0.360Improved
add LLDS-MA (answer-token masking) (ablation)GRPO, Qwen2.5-3B-Base · 7 QA benchmarks avgEM: 0.360 -> 0.440 (+45.2% over GRPO)Improved
add LLDSGRPO, Qwen2.5-3B-Instruct · 7 QA benchmarks avgEM: 0.336 -> 0.441Improved
add LLDSGRPO, Qwen2.5-7B-Base · 7 QA benchmarks avgEM: 0.350 -> 0.480Improved
add LLDSGRPO, Qwen2.5-7B-Instruct · 7 QA benchmarks avgEM: 0.396 -> 0.483Improved
add LLDS to GSPOGSPO, Qwen2.5-3B-Instruct · 7 QA benchmarks avgEM: 0.331 -> 0.451Improved
regularization weight lambda 0.01 instead of 0 (ablation)GRPO + LLDS, Qwen2.5-3B-Instruct · training rewardcollapse step: collapse ~200 -> ~220Mixed
regularization weight lambda 0.2 instead of 0 (ablation)GRPO + LLDS, Qwen2.5-3B-Instruct · training rewardtraining stabilityImproved
use LLDS instead of dense-reward Tree-GRPOGRPO, Qwen2.5-7B · 7 QA benchmarks avgEM: 0.439 -> 0.480Improved
add LLDS across model sizes and variantsGRPO · training reward curvescollapseImproved

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