TY  - RPRT
TI  - Approximation of Log-Partition Function in Policy Mirror Descent Induces Implicit Regularization for LLM Post-Training
AU  - Zhenghao Xu
AU  - Qin Lu
AU  - Changlong Yu
AU  - Tuo Zhao
PY  - 2026
UR  - https://arxiv.org/abs/2602.05933
ID  - 2602.05933
ER  -