非負精緻化をともなうPrivelet法の演算効率化手法

IPSJ

Bibliographic Information

Other Title
  • A Method to Improve the Computational Efficiency of Privelet with Nonnegative Refinement

Search this article

Abstract

Privelet法は，差分プライバシ基準に準拠しつつ，部分和精度にも優れており，プライバシが保護されたデータのスケーラブルな活用を可能にする．このPrivelet法に非負精緻化処理を組み込むと，高い部分和精度を維持しつつ，さらに，「非負制約の逸脱」や「疎データの密度急増」という2つの問題への対処も可能となる．この手法の場合，非負精緻化をともなう逆Wavelet変換（Top-down精緻化）部分に枝刈り処理を導入することで，演算を効率化することができる．しかし，その効果の程度，性質などについては，いまだ明らかにされていない．そこで本稿では，演算効率化が期待できる実装法を用いて，枝刈り処理の性能面に対する評価を行い，さらにその特性についての考察を行う．

Privelet is a data publishing technique that ensure ε-differential privacy while providing accurate answers for range-count queries. This technique is suitable for scalable utilization of privacy-preserved data. However, it has two problems which are “deviation from the non-negative constraint” and “abruptly increase of data-density”. Privelet with non-negative refinement solves these two problems without losing the accuracy of the partial summation. Introducing the pruning process into the top-down refinement - the inverse wavelet transform with nonnegative refinement - improves the computational efficiency. However, its effects and characteristics have not been clarified yet. In this paper, after showing the evaluation results for performance using an implementation method which can be expected to improve the efficiency of computation, its characteristics are discussed.

Journal

Keywords

Details 詳細情報について

Report a problem

Back to top