FRIARS: a feedback control system for information assurance using a Markov decision process

The authors describe a prototype cyber-defense decision system for autonomously reacting to system attacks by using a Markov Decision Process (MDP), a simple game, and a weighted random culling from select actuators. The system is novel in that the amount of knowledge engineering to build the system is minimal, consisting of knowing the approximate false alarm rate of each of the sensors, the probability of failure for each of the actuators, and an estimate of cost associated with each actuator. A system of sensors and actuators can thus be built a priori and then deployed at the host or network level