Autonomous Cyber Security
Abstract
We describe a set of decision-making algorithms (centralized, decentralized, and hybrid) for a multi agent system and present the Factored-Value (FV) Monte Carlo Tree Search (MCTS) Hybrid Cost Max-Plus algorithm for Autonomous Cyber Security. Our proposed algorithm consists of a two-step process. In the first step, each agent uses MCTS to find its best individual actions, taking into account cost. Each agent presents its most promising actions to the team. In the second step, the Hybrid Cost Max-Plus algorithm is used for joint action selection. This Hybrid Cost Max-Plus algorithm improves upon the known centralized and distributed Max-Plus algorithms without cost by including the cost of actions in the interactions of agents. The Max-Plus algorithm uses the framework of Coordination Graphs, which exploit dependencies among agents to decompose the global payoff function as the sum of local terms. Our proposed FV-MCTS-Hybrid-Cost-Max-Plus algorithm is online, anytime, distributed, and scalable in terms of the number of agents and their interactions. Our contribution competes with state of art methods and algorithms that use MCTS and Max-Plus to exploit the locality of agent interactions for planning and acting.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 12, 2023
- Accession Number
- AD1210468
Entities
People
- Alexander Velazquez
- Myong Kang
Organizations
- United States Naval Research Laboratory