Exclusive Strategy for Generalization Algorithms in Micro-data Disclosure

Abstract

Abstract. When generalization algorithms are known to the public, an adversary can obtain a more precise estimation of the secret table than what can be deduced from the disclosed generalization result. Therefore, whether a generalization algorithm can satisfy a privacy property should be judged based on such an estimation. In this paper, we show that the computation of the estimation is inherently a recursive process that exhibits a high complexity when generalization algorithms take a straightforward inclusive strategy. To facilitate the design of more efficient generalization algorithms, we suggest an alternative exclusive strategy, which adopts a seemingly drastic approach to eliminate the need for recursion. Surprisingly, the data utility of the two strategies are actually not comparable and the exclusive strategy can provide better data utility in certain cases.

Preserving Privacy Against Side-Channel Leaks

“…To improve the efficiency, a so-called exclusive strategy is proposed in [112] to penalize the cases where a recursive process is required to compute the adversarial mental image about the micro-data table. To examine the general case, we have proposed a kjump strategy [73](see Chapter 3 for the first line of our research) to penalize such cases where with more control in the sense that only k, instead of all, generalization functions will be skipped.…”

k-jump: A strategy to design publicly-known algorithms for privacy preserving micro-data disclosure

“…Our proposed family of algorithms is general to handle different syntactic privacy properties and different measures of data utility. Closest to this work, a special case of the k-jump strategy is discussed in [51] where all jumps end at disclosing nothing. Our result in this paper is more general than those in [51].…”

“…Closest to this work, a special case of the k-jump strategy is discussed in [51] where all jumps end at disclosing nothing. Our result in this paper is more general than those in [51]. It is also worth noting that we substantially extend our previous version [35] by elaborating on the proofs, analyzing the computational complexity, and confirming the necessity of safe algorithms even when making the choice secret among the algorithms.…”