@article{tamingextremetokenscovarianceawaregrpowi, title = {Taming Extreme Tokens: Covariance-Aware GRPO with Gaussian-Kernel Advantage Reweighting}, author = {Cheng Wang and Qin Liu and Wenxuan Zhou and Muhao Chen}, year = {2026}, eprint = {2605.11538}, archivePrefix = {arXiv}, url = {https://arxiv.org/abs/2605.11538}, }