Багатоагентне навчання з підкріпленням для виявлення вторгнень

Abstract

This paper explores the application of multi-agent reinforcement learning (MARL) for intrusion detection in modern network environments. The increasing complexity and distributed nature of information systems require adaptive and intelligent security solutions capable of detecting both known and unknown cyber threats. The proposed approach utilizes multiple cooperative agents that analyze network traffic and system behavior in parallel, learning optimal detection strategies through interaction with the environment. Special attention is given to coordination mechanisms between agents and the use of deep reinforcement learning methods to improve detection accuracy. The advantages of MARL-based systems, including scalability, adaptability, and robustness, are discussed alongside challenges such as training instability, computational complexity, and data requirements. The study demonstrates that multi-agent reinforcement learning can significantly enhance the effectiveness of intrusion detection systems in dynamic and large-scale network infrastructures