L2RAVE: Feedback-Driven Learn to Reason in Adversarial Environments for Autonomic Cyber Systems
Abstract
The growing complexity of cyber systems has made them difficult for human operators to defend, particularly in the presence of intel""ligent and resourceful adversaries who target multiple system components simultaneously, employ previously unobserved attack vectors"", and use stealth and deception to evade detection. There is a need for developing autonomic cyber systems that can integrate statis""tical learning and rules-based formal reasoning to provide an adaptive and robust situational awareness, decision making, planning a""nd executing actions.In this collaborative research effort, we propose to develop a feedback-driven Learn to Reason (L2R) framewor""k, which aims to integrate statistical learning with formal reasoning, in adversarial environments. Our insight is that in order to"" realize the potential benefits of L2R, continuous interaction between the statistical and formal components is needed, both at inte"rmediate time steps and at multiple layers of abstraction. The proposal addresses the following research problems.Statistical and Formal Reasoning Models: In this research thrust we will investigate and develop statistical and formal reasoning models that accurately learn the adversary~s behav-iors over time. We will be addressing the research questions: (i) How to learn the temporal dependencies of the adversary behaviors and correlated adversary actions over time? (ii) How to provide accurate learning in the presence of adaptive attackers? We will investigate and develop the formal reasoning model via the research questions: (iv) How to formalize the concept of rules in a human interpretable and actionable manner? (v) How to translate human domain knowledge into machine-understandable rules? and (vi) How to incorporate uncertainties in the rules?Enhancement of Statistical Learning via Formal Reasoning: This thrust will investigate how to improve the accuracy and interpretability of statistical learning through formal reasoning. The following research questions arise: (i) How to generate data samples from a given set of rules that are maximally beneficial to the" learning process? (ii) For rules that are uncertain, how to incorporate the uncertainty when learning from samples generated by unc"ertain rules? (iii) How to translate the formal rules into a language that is understandable by intermediate layers of the statistical model? (iv)How to influence learning at the inter-mediate layers based on the rules? and (v) How to ensure that the system output agrees with both the learned statistical representation of the data and the formal rules?Thrust Three ~ Statistical Learning-Aided Rule Update: The rules governing system and adversary behavior willnaturally evolve over time due to changes in the environment and adaptive attack strategies. We will address the following research questions: (i) How to update the levels of uncertainty for each rule based on the statistical model? (ii) How to remove or revise rules that are contradicted by the observed data? (iii) How to enhance the interpretability of the intermediate outputs of the statistical model? (iv) How to automatically propose new rules based on commonly occurring features from the statistical model? and (v) How to incorporate side information into the formal reasoning?
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 29, 2017
- Source ID
- N000141712946
Entities
People
- Radha Poovendran
Organizations
- Office of Naval Research
- United States Navy
- University of Washington