Computational improvements in parallelized k-anonymous microaggregation of large databases

Abstract

The technical contents of this paper fall within the field of statistical disclosure control (SDC), which concerns the postprocessing of the demographic portion of the statistical results of surveys containing sensitive personal information, in order to effectively safeguard the anonymity of the participating respondents. The concrete purpose of this study is to improve the efficiency of a widely used algorithm for k-anonymous microaggregation, known as maximum distance to average vector (MDAV), to vastly accelerate its execution without affecting its excellent functional performance with respect to competing methods. The improvements put forth in this paper encompass algebraic modifications and the use of the basic linear algebra subprograms (BLAS) library, for the efficient parallel computation of MDAV on CPU.