CTA is a post-tabular perturbative approach for statistical disclosure control. Its purpose is to compute the closest safe table to the original data, using some distance. Sensitive cells are adjusted either upwards or downwards (binary decision), and the resulting cells have to be accordingly (and minimally) modi_ed to preserve marginals. For real and large tables, CTA may result in a dicult mixed integer linear problem for some weights in the objective function. In those situations the Block Coordinate Descent (BCD) heuristic for CTA|which is included in the Tau-Argus CTA distribution|may be used to quickly obtain a feasible, hopefully close to optimality, solution. We present a practical experiment using a large and di_cult real-world table from Eurostat. We will show that, for unitary weights, while the standard CTA can not obtain a solution in about half an hour, the BCD-CTA approach provides a solution in few seconds.

Peer Reviewed