Domain reduction dependencies: A new type of dependency for statistical database security

Abstract

A new type of data dependency, the domain reduction dependency is defined. Given a set of attributes over which a query is being made, this dependency allows one to more accurately predict the probability of release of sensitive statistics for that query than could have been done with table size alone. In addition, the properties of DRDs provide metainformation on the gain and loss of information within a lattice of queries through the addition or removal of attributes from a query set. This increased formalism on information loss and gain may provide insights which will allow the construction of better techniques for protecting the security of future statistical databases.<>