AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Lookahead Tree-Based Rollouts for Enhanced Trajectory-Level Exploration in Reinforcement Learning with Verifiable Rewards

Shangyu Xing, Siyuan Wang, Chenyuan Yang, Xinyu Dai, Xiang Ren

ICLR 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
replace stochastic sampling rollouts with LATRDAPO (Qwen2.5-3B) · CountdownPass@1: 70.7 -> 74.7Improved
replace stochastic sampling rollouts with LATRGRPO (Qwen2.5-3B) · CountdownPass@1: 65.9 -> 70.9Improved
replace stochastic sampling rollouts with LATRDAPO (Qwen2.5-3B) · AMC-2023Pass@1: 37.8 -> 45.3Improved
replace stochastic sampling rollouts with LATRDAPO (Qwen2.5-3B) · MATH-500Pass@8: 79.2 -> 79.0No clear effect
replace stochastic sampling rollouts with LATRDAPO (Qwen2.5-3B) · Countdownaverage completion length: 483 -> 367Improved
use LATR on larger/other base modelsQwen2.5-7B · CountdownPass@1: 70.9 -> 76.0Improved
use random branching instead of probability-threshold branching (ablation)LATR · Countdown (DAPO)Pass@1: 69.6 -> 74.7Improved
add similarity-based pruning (ablation)LATR · Countdown (DAPO)Pass@1: 71.0 -> 74.7Improved
use trajectory-level lookahead instead of token-level lookahead (ablation)LATR · Countdown (DAPO)Pass@1: 72.1 -> 74.7Improved
use embedding-based similarity for pruning instead of edit distance (ablation)LATR · Countdown (DAPO)Pass@1: 74.7 -> 72.9Hurt
use ROUGE-L or suffix matching instead of edit distance (ablation)LATR · Countdown (DAPO)Pass@1: 74.7 vs 73.9 (ROUGE-L), 74.9 (suffix)No clear effect
use rollout down-sampling (RDS) with DAPO instead of stochastic (ablation)DAPO · CountdownPass@1: 70.7 -> 68.5Hurt

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