📄 中文摘要
揭示因果结构对于理解复杂系统和做出明智决策至关重要。强化学习（RL）在识别这些结构方面显示出潜力，尤其是以有向无环图（DAG）的形式。然而，现有方法往往效率低下，不适合在线应用。MARLIN是一种高效的基于多智能体的强化学习方法，旨在增量学习DAG。该方法采用DAG生成策略，将连续实值空间映射到DAG空间，作为内部批处理策略。同时，MARLIN结合了两种状态特定和状态不变的RL智能体，以揭示因果关系，并将这些智能体整合到增量学习框架中。
MARLIN：用于增量DAG发现的多智能体强化学习
出处: MARLIN: Multi-Agent Reinforcement Learning for Incremental DAG Discovery
发布: 2026年3月24日
📄 English Summary
MARLIN: Multi-Agent Reinforcement Learning for Incremental DAG Discovery
Uncovering causal structures is essential for understanding complex systems and making informed decisions. Reinforcement learning (RL) has shown potential in identifying these structures in the form of directed acyclic graphs (DAGs). However, existing methods often lack efficiency, rendering them unsuitable for online applications. MARLIN is proposed as an efficient multi-agent RL approach for incremental DAG learning. It employs a DAG generation policy that maps a continuous real-valued space to the DAG space as an intra-batch strategy. Additionally, MARLIN integrates two types of RL agents—state-specific and state-invariant—to uncover causal relationships and incorporates these agents into an incremental learning framework.