Preservation of probabilistic information flow under refinement

Abstract

Information flow properties, which describe confidentiality requirements, are not generally preserved under behavior refinement. This article describes a formal framework for refinement relations between nondeterministic probabilistic processes that capture sufficient conditions to preserve information flow properties. In particular, it uses information-theoretic concepts to investigate the refinement of a probabilistic, entropy-based information flow property. The refinement relation considers the abstract and concrete models as views on the same stochastic process. Probabilistic CSP provides the semantic basis for this investigation.