AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Stable Asynchrony: Variance-Controlled Off-Policy RL for LLMs

Luke J. Huang, Zhuoyang Zhang, Qinghao Hu, Shang Yang, Song Han

ICML 2026 · Reinforcement learning › RL from feedback · approach: RL for LLM reasoning

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 ESS-guided step-size scaling and off-policy optimal baseline to TIS (ablation)sequence-level TIS · MATH-500, Qwen2.5-7B, 10 steps off-policystability / accuracyImproved
add ESS-guided learning-rate scaling to TIS (ablation)TIS · MATH-500, Qwen2.5-7Bstability / accuracyImproved
add off-policy optimal baseline (OPOB) to TIS (ablation)TIS · MATH-500, Qwen2.5-7Bstability / accuracyImproved
use VCPO instead of sequence-level TISTIS · GSM8K, Qwen2-1.5B, 12 steps off-policytraining stabilityImproved
use VCPO with async k=10 instead of synchronousSync RL · Countdown, Qwen2.5-7BGPU hours: 143.2 -> 89.6Improved
use VCPO with async k=2 instead of synchronousSync RL · AIME 2025 tool-use, Qwen2.5-7BGPU hours: 420.2 -> 168.9Improved
increase policy lag up to 128 steps off-policy with VCPO (ablation)VCPO · Qwen2.5-7B mathstabilityNo clear effect
lower the learning rate instead of ESS scaling (ablation)TIS · MATH-500, high asynchronyaccuracyHurt
raise TIS threshold from c=2.0 to c=8.0 (ablation)sequence-level TIS · GSM8K, Qwen2-1.5Bcollapse delayImproved
use sequence-level MIS masking (ablation)MIS · GSM8K, Qwen2-1.5BstabilityHurt
use M2PO-style masking (max 0.04) (ablation)M2PO · GSM8K, Qwen2-1.5B, k=12stabilityHurt
add KL-in-reward (beta=0.001) (ablation)Seq-TIS · GSM8K, Qwen2-1.5B, k=12validation accuracyHurt

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