Reinforcement learning (RL) has emerged as a key method for training AI models in scientific applications, but RL remains difficult when outputs cannot easil…
机构：UCSD
来源：arXiv 2610.25625 | AI4Papers 论文推荐平台