We are looking to understand the decision processes and factors cyber attackers use to achieve their goals so that defensive measures can be deployed. The complexity of network infrastructure, defensive policies, and the secretive nature of cyber attacks make assessing the behaviors difficult. We hope to explore ML techniques like RL to assist in understanding the different cyber-attack processes. Reinforcement learning leveraging cyber-attack simulation enables us to search for rewarding attacker campaigns; however, this approach has several downsides. This is why we will be exploring the concept of Neuro-Symbolic learning to develop human-interpretable models of deep decision processes so that mitigation efforts can be linked to a specific attacker’s decision or behavior.

I am participating in SuperUROP because I want to gain high-level research experience in ML & AI. I teach the MIT Pokerbots competition every year, where we teach some of these concepts to the students. With this SuperUROP, I hope to further my knowledge of ML & AI, especially in the context of multi-agent systems.