Exploitation as an inference problem

Abstract

In this position paper, we suggest that an adversary seeking to exploit a side channel should be viewed as performing inference under uncertainty. We propose a set of vulnerability measures that incorporate both observational effort and computational effort. by deriving Boolean satisfiability as a special case of the marginalization problem, we justify that the measure is capable of capturing the complexity of the underlying deterministic decision problem. In the limit of unbounded computation the measure reduces to the efficiency (in number of observations) of naïve Bayesian analysis. We further hypothesize that in the limit of unbounded observations, the measure reduces to the complexity of the decision problem.