Deep reinforcement learning for optimal denial-of-service attacks scheduling

We consider an optimal denial-of-service (DoS) attack scheduling problem of N independent linear time-invariant processes, where sensors have limited computational capability. Sensors transmit measurements to the remote estimator via a communication channel that is exposed to DoS attackers. However, due to limited energy, an attacker can only attack a subset of sensors at each time step. To maximally degrade the estimation performance, a DoS attacker needs to determine which sensors to attack at each time step. In this context, a deep reinforcement learning (DRL) algorithm, which combines Q-learning with a deep neural network, is introduced to solve the Markov decision process (MDP). The DoS attack scheduling optimization problem is formulated as an MDP that is solved by the DRL algorithm. A numerical example is provided to illustrate the efficiency of the optimal DoS attack scheduling scheme using the DRL algorithm.