Network_Defense_Symmetric_Competitive
收藏资源简介:
该数据集是一个专注于多智能体强化学习(MARL)的模拟环境,包含102,400,000个时间步。数据集模拟了网络安全领域的红蓝对抗场景:红方智能体扮演攻击者,其目标遵循MITRE ATT&CK框架,包括发现漏洞、提升权限、破坏资产和维持持久性;蓝方智能体扮演防御者,其目标包括维护系统可用性、减少攻击面、检测恶意行为、使用蜜罐欺骗攻击者以及驱逐入侵者,行动空间涉及防御配置和缓解步骤。红蓝双方为零和对抗关系,即一方的收益完全等于另一方的损失。该数据集适用于训练和评估网络安全领域的多智能体强化学习算法,特别侧重于攻击与防御的动态博弈。
This dataset is a simulation environment focused on multi-agent reinforcement learning (MARL), containing 102,400,000 timesteps. It simulates a red team-blue team adversarial scenario in the cybersecurity domain: the red agents act as attackers with objectives aligned with the MITRE ATT&CK framework, including discovering vulnerabilities, escalating privileges, compromising assets, and maintaining persistence; the blue agents act as defenders with objectives such as maintaining system availability, reducing attack surfaces, detecting malicious behavior, using honeypots to deceive attackers, and expelling intruders, with action spaces involving defense configurations and mitigation steps. The red and blue teams are in a zero-sum adversarial relationship, meaning one sides gain equals the others loss. The dataset is suitable for training and evaluating MARL algorithms in cybersecurity, with a particular focus on the dynamic game of attack and defense.
数据集概述:Network Defense Symmetric Competitive
- 数据集名称:Network Defense Symmetric Competitive
- 托管平台:Hugging Face
- 数据集规模:约 102,400,000 时间步(属于 100M < n < 1B 规模类别)
- 语言:英语(en)
- 许可证:MIT
核心内容
-
任务类型:多智能体强化学习(Multi-Agent Reinforcement Learning)
-
场景设定:红蓝对抗的纯零和博弈(Pure Zero-Sum Opposition)
- 红方(Red Agent):目标是发现系统漏洞、提升权限、破坏资产并维持持久化控制。其动作空间可参照 MITRE ATT&CK 框架的各阶段建模。
- 蓝方(Blue Agent):目标是维护系统可用性、减小攻击面、检测恶意行为、利用蜜罐欺骗攻击者并驱逐入侵者。其动作空间包含防御配置和缓解措施。
-
博弈性质:红方的收益即为蓝方的绝对损失,反之亦然,构成完全对称的零和竞争关系。
相关资源
- 防御智能体演示视频:YouTube 链接




