K-SNCC: group deviations in subsidized non-cooperative computing

Abstract

A function is subsidized non-cooperative computable [SNCC] if honest agents can compute it by reporting truthfully their private inputs, while unilateral deviations by the players are not beneficial: if a deviation from truth revelation can mislead other agents, this deviation will decrease the deviator's chances of correct computation, or, it will not affect these chances but the expected payment to the deviator will decrease; in addition, deviations can not increase the expected monetary payments to a deviator without decreasing his chances of correct computation. This paper extends the study of SNCC functions to the context of group deviations. A function is K-SNCC if deviations by a group of at most K agents are not beneficial. We provide a full characterization of the K-SNCC functions, both for the independent values and the correlated values settings.