Theoretical Foundations of Adversarial Binary Detection

The present monograph focuses on the detection problem in adversarial setting. When framed in an adversarial setting, classical detection theory can not be applied any more, since, in order to make a correct decision, the presence of an adversary must be taken into account when designing the detector. In particular, the interplay between the Defender (D), wishing to carry out the detection task, and the Attacker (A ), aiming at impeding it, must be investigated. The purpose of this monograph is to lay out the foundations of a general theory of adversarial detection, taking into account the impact that the presence of the adversary has on the design of the optimal detector. We do so by casting the adversarial detection problem into a game theoretical framework, which is then studied by relying on typical methods of information theory. As a final result, the theory allows to state the conditions under which both the false positive and false negative error probabilities tend to zero exponentially fast, and to relate the error exponents of the two kinds of errors to the distortion the attacker can introduce into the test sequence.