TY - CONF
AU - Baena, D.
AU - Castro, J.
AU - Frangioni, A.
T3 - Joint UNECE/Eurostat Work Session on Statistical Data Confidentiality 2019
PY - 2019
Y1 - 2019
SP - 1
EP - 3
PB - John Wiley & sons
UR - https://statswiki.unece.org/display/confid/Work+Session+on+Statistical+Data+Confidentiality+2019
AB - The cell suppression problem (CSP) is one of the most widely applied post-tabular protection methods for tabular data. Given a set of sensitive cells to be protected, its aim is to find set of additional cells whose removal guarantees that estimates of values of sensitive cells fall out of a predefined protection interval. CSP is formulated as a very large mixed integer linear optimization problem. Due to its large scale is a very challenging problem for current general optimization solvers, and specialized approaches (namely, cutting planes or Benders decomposition) are needed for its solution (for instance, Benders decomposition is the algorithm used in the state-of-the-art optimal CSP implementation in Tau-Argus). However, the convergence to the optimal solution is often too slow due to well known instability issues of Benders decomposition. This work discusses a recently developed stabilized Benders method, in an attempt to avoid some of its instabilities. Some results are reported in the solution of generated and real-world CSP instances, showing the effectiveness of this approach. In some instances, stabilized Benders provided a very good solution in less than one minute, while the other available approaches (i.e., general or specialized codes, such as CPLEX or the optimal option in Tau-Argus) found no feasible solution in one hour.
T2 - Joint UNECE/Eurostat Work Session on Statistical Data Confidentiality
TI - Using a stabilized Benders algorithm for cell suppression
ER -