TY  - RPRT
TI  - Transformers Provably Implement In-Context Reinforcement Learning with Policy Improvement
AU  - Haodong Liang
AU  - Lifeng Lai
PY  - 2026
UR  - https://arxiv.org/abs/2605.05755
ID  - 2605.05755
ER  -