Gradient Starvation in Binary-Reward GRPO: Why Group-Mean Centering Fails and Why the Simplest Fix Works

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Nie, Wenhua, Wu, Jianan, Liu, Junlin, Li, Ziwei, Lin, Zheng, Zijian, Zhang, Fan, Yilong, Zheng, Haoran, Jang, Jyh-Shing Roger
Format:	Preprint
Published:	2026
Subjects:	Machine Learning
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Internet

https://arxiv.org/abs/2605.07689