カンファレンス (国際) Mitigating Privacy Vulnerability Caused by Map Asymmetry

Ryota Hiraishi (Kyoto univ.), Masatoshi Yoshikawa (Kyoto univ.), Shun Takagi (Kyoto univ.), Yang Cao (Kyoto univ.), Sumio Fujita, Hidehito Gomi

36th Annual IFIP WG 11.3 Conference on Data and Applications Security and Privacy (DBSEC 2022)


Recently, Geo-Indistinguishability (GeoI) has been researched as a privacy concept to protect location information in location based services (LBSs). GeoI is defined as an extension of differential privacy, so it is a mathematically rigorous concept of location privacy. However, the Laplace mechanism that satisfies GeoI outputs location information equally in all directions because the map is considered to be symmetric. Therefore, when noise is added, the output can be directed to locations that are considered uninhabitable, such as the sea. To solve this problem, a study has proposed a graph representation of the road network, Geo-Graph-Indistinguishability (GeoGI), which is differential privacy on the graph, and a Graph-Exponential Mechanism (GEM), which satisfies the GeoGI. However, the problem that some vertices can be vulnerable to privacy when measured using criteria based on adversary’s error remains unresolved. In this paper, we first calculate the degree of privacy protection using the adversary’s error when noise is added by GEM, and show that privacy is weakened at some vertices. Next, we propose two methods to mitigate this problem, weight reduction method and AE optimization method, and show through experiments that the application of these methods actually mitigates the privacy vulnerability. Finally, we point out the problems with these two methods and suggest directions for their solutions.

Paper : Mitigating Privacy Vulnerability Caused by Map Asymmetry (外部サイト)