PROJECT TITLE :

A Game-Theoretic Analysis of Adversarial Classification - 2017

ABSTRACT:

Attack detection is typically approached as a classification drawback. However, commonplace classification tools often perform poorly, because an adaptive attacker will shape his attacks in response to the algorithm. This has led to the recent interest in developing methods for adversarial classification, however to the simplest of our knowledge, there are a terribly few previous studies that take under consideration the attacker's tradeoff between adapting to the classifier getting used against him together with his want to take care of the efficacy of his attack. Together with this effect may be a key to derive solutions that perform well in observe. In this investigation, we tend to model the interaction as a game between a defender who chooses a classifier to differentiate between attacks and traditional behavior based mostly on a set of observed options and an attacker who chooses his attack options (class one information). Traditional behavior (category zero data) is random and exogenous. The attacker's objective balances the benefit from attacks and the value of being detected whereas the defender's objective balances the advantage of a correct attack detection and the cost of false alarm. We give an efficient algorithm to compute all Nash equilibria and a compact characterization of the possible varieties of a Nash equilibrium that reveals intuitive messages on how to perform classification within the presence of an attacker. We additionally explore qualitatively and quantitatively the impact of the non-attacker and underlying parameters on the equilibrium strategies.