Adversarial Behavior in Network Games

Abstract

This paper studies the effects of and countermeasures against adversarial behavior in network resource allocation mechanisms such as auctions and pricing schemes. It models the heterogeneous behavior of users, which ranges from altruistic to selfish and to malicious, within the analytical framework of game theory. A mechanism design approach is adopted to quantify the effect of adversarial behavior, which ranges from extreme selfishness to destructive maliciousness. First, the well-known result on the Vicrey-Clarke-Groves (VCG) mechanism losing its efficiency property in the presence of malicious users is extended to the case of divisible resource allocation to motivate the need to quantify the effect of malicious behavior. Then, the Price of Malice of the VCG mechanism and of some other network mechanisms are derived. In this context, the dynamics and convergence properties of an iterative distributed pricing algorithm are analyzed. The resistance of a mechanism to collusions is investigated next, and the effect of collusion of some malicious users is quantified. Subsequently, the assumption that the malicious user has information about the utility function of selfish users is relaxed, and a regression-based iterative learning scheme is presented and applied to both pricing and auction mechanisms. Differentiated pricing as a method to counter adversarial behaviors is proposed and briefly discussed. The results obtained are illustrated with numerical examples and simulations.

Bayesian mechanisms for wireless network security

“…The model and assumptions which we present in this section are based on our earlier paper, [8], and is partially repeated here for completeness. At the center of the mechanism design model is the designer D who influences N users, denoted by the set A.…”

“…To quantify the effect of the malicious behavior a metric, P oM, was proposed in [8] which is defined as,…”

“…The power strategies of the Regular user and the User of Unknown Type at the Bayesian NE point for a game between two users of logarithmic utility function given in (8) are given by,…”

“…whether others are regular users or botnets (jammers). We extend the work in [8] to an incomplete information case where the malicious users are countered without explicit detection or learning. We assume only that the designer knows the probability of malicious users' existence and he counters them by updating the prices using the probabilistic information.…”

“…We also use a different model of the malicious behavior in this paper compared to the one in [9]. The new model is proposed in our prior work in [8], where the malicious user is interested in his own utility and has a degree of maliciousness. Our aim is to obtain prices in a Kelly-type mechanism [11], for interference coupled networks in the presence of malicious users.…”