Numerical computation of multi-goal security strategies

Abstract

Security is often investigated in terms of a single goal (e.g., confidentiality), but in practical settings mostly a compound property comprising multiple and often interdependent aspects. Security strategies are behavior profiles that guarantee some performance regardless of how the adversary really behaves (provided that it stays within its action set). While security strategies towards a single goal are easy to compute via Nash-equilibria (or refinements thereof), playing safe towards multiple security goals induces the notion of Pareto-optimal security strategies. These were recently characterized via Nash-equilibria of multi-player games, for which solution algorithms are intricate and may fail for small instances already. Iterative techniques, however, exhibited good stability even for large games. In this work, we thus report on theoretical and practical results how security strategies for multiple (interdependent) goals can be computed via a set of simple transformations and a final application of humble fictitious play.