强化学习策略验证的综述
A Survey on the Verification of Reinforcement Learning Policies
浏览论文内容
中文总结 AI 辅助
综述强化学习策略验证，引入分类法沿验证范式、时间范围和保证强度三个轴阐明现有方法关系，统一理论基础，明确假设局限，确定新兴方向，为RL验证提供统一视角。
中文摘要 AI 辅助
强化学习（RL）在复杂的安全关键领域应用日益广泛，但基于神经网络的策略缺乏严格行为保证仍是部署的主要障碍。政策表达性和规模的最新进展加剧了这一挑战，导致RL政策验证工作迅速增长但概念上分散。本综述提供了RL验证方法的统一视角。我们引入一种分类法，沿验证范式（形式与概率）、时间范围（逐步与多步）和保证强度三个轴阐明现有方法之间的关系。除了分类法，我们统一了基础理论基础，明确了隐含假设和局限性，并确定了新兴方向。
英文摘要
Reinforcement learning (RL) is increasingly applied in complex, safety-critical domains, yet the lack of rigorous behavioral guarantees for neural network-based policies remains a major barrier to deployment. Recent advances in policy expressiveness and scale have intensified this challenge, leading to a rapidly growing but conceptually fragmented body of work on RL policy verification. This survey provides a unifying perspective on RL verification methods. We introduce a taxonomy that clarifies relationships among existing approaches along three axes: verification paradigm (formal versus probabilistic), temporal scope (step-wise versus multi-step), and guarantees strength. Beyond taxonomy, we unify underlying theoretical foundations, make implicit assumptions and limitations explicit, and identify emerging directions.
发表机构
- TU Wien（维也纳工业大学）
- Massachusetts Institute of Technology（麻省理工学院）
机构由 AI 辅助整理，请以论文原文为准。
