Multi-objective Learning Techniques—Secure Hypergraph Informed Robust Reinforcement Learning for Autonomous Cyber Defense (MOLT–SHIELD)

PIs: Sam Chatterjee and Emilie Purvine

MOLT–SHIELD incorporated complex state representations into multiagent systems for cyber defense. Specifically, we developed a context-aware large language model–guided reward design approach for multiple heterogeneous attack and defense agents within an autonomous deep reinforcement learning–based cyber defense simulation environment with a multihost cyber network. Our work was accepted and presented at the Association for the Advancement of Artificial Intelligence (AAAI)-26 AI for Cyber Security Workshop (https://arxiv.org/pdf/2511.16483).

PNNL

Generative AI

Multi-objective Learning Techniques—Secure Hypergraph Informed Robust Reinforcement Learning for Autonomous Cyber Defense (MOLT–SHIELD)