Learning Algorithms for Autonomous Security in (Mixed-)Autonomous Networks
Abstract
Today’s cyber and physical networks are poised to become even more autonomous in the future. Driverless transportation is likely to be part of the options. Wars may be fought autonomously. Flexible energy systems with mobile generators may be able to autonomously mobilize and respond to disruptions in real time. The scale, scope and pace of autonomous systems – driven by rapid advances in AI-ML-driven platforms, powerful edge devices (both sensors and actuators) and communication networks – are steadily outperforming humans and human-controlled systems. Many of these capabilities are also making attacks more advanced (due to the availability of sophisticated tools to bypass off-the-shelf defense schemes), targeted (since attackers seek to gain knowledge about the system and the component valuations to position their attacks), persistent (in achieving the adversarial objective by being insistent yet selective in launching the attacks), adaptive (by learning defense strategies during interactions with the defender and tailoring strategies), and deceptive (concealing the true intention and using disguise to evade detection). Thus, it is clear that future autonomous systems deployed in highly contested environments will require new cyber defense strategies and tactics, and associated tools and procedures. This project is motivated by the basic challenge in the development and automation of cyber defense in complex networks that are prone to autonomous adversaries. Research Focus. We focus on developing theoretical and computational foundations for a class of adaptive online learning algorithms that will facilitate- (1) Reasoning about situation awareness and dynamic alert prioritization based on noisy and potentially compromised observations from network sensors (2) Cyber deception to provide misleading information and shape their perception to make them erroneously confident and waste their time-resources by conducting poor actions; (3) Response operations and decisions in autonomous cyber-physical systems, given that adversarial capabilities will also change with these developments.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 29, 2024
- Source ID
- FA95502310190
Entities
People
- Patrick Jaillet
Organizations
- Air Force Office of Scientific Research
- Massachusetts Institute of Technology
- United States Air Force