Privacy Analysis with a Distributed Transition System and a Data-Wise Metric
Résumé
We introduce a logical framework DLTTS (Distributed Labeled Tagged Transition System), built using concepts from Proba- bilistic Automata, Probabilistic Concurrent Systems, and Probabilistic labelled transition systems. We show that DLTTS can be used to formally model how a given piece of private information P (e.g. a tuple) stored in a given database D protected by generalization and/or noise addition mechanisms, can get captured progressively by an agent repeat- edly querying D, by using additional non-private data, as well as knowl- edge deducible with a more general notion of adjacency based on metrics defined ‘value-wise’; such metrics also play a role in differentially private protection mechanisms.