TY  - RPRT
TI  - Mitigating Shortcut Reasoning in Language Models: A Gradient-Aware Training Approach
AU  - Hongyu Cao
AU  - Kunpeng Liu
AU  - Dongjie Wang
AU  - Yanjie Fu
PY  - 2026
UR  - https://arxiv.org/abs/2603.20899
ID  - 2603.20899
ER  -