# Regularized Score Centering

> Source: <https://aiflash.com/news/153534/>
> Published: 2026-11-18 04:30:41+00:00

**Recommendation:** When asynchronous RL training introduces stale-policy data and train/inference numerical mismatch, apply score centering to strip reward-baseline effects from the expected update and stabilize off-policy learning.
More in the playbook "Fine-tuning and data": <https://wpnews.pro/playbooks/training-data>

Asynchronous reinforcement learning increases training throughput, but data from stale policies and numerical mismatch between training and inference can destabilize learning. To stabilize off-policy learning, score centering (SC) removes the effect of reward baselines on the expected update. Our gr
