control-policies

Tag

#control-policies

ArcFlow: Training Flow Policies for Multiple Solver Budgets

arXiv cs.LG ↗ · 2026-10-22 Cached

ArcFlow 提出一种部署预算感知的训练算法，使单个流策略（flow policy）能在多个求解器步数预算下保持高性能，尤其在一步推理场景中优于五个基线方法。

0 favorites 0 likes