Enforcing Adaptive Location Privacy with Federated Learning

Yanis Meziani (1, 2, 3, 4) , Besma Khalfoun (2, 3, 4) , Sara Bouchenak (2, 3, 4) , Sonia Ben Mokhtar (2, 3, 4) , Vlad Nitu (2, 3, 4)

1 ESI - École Nationale Supérieure d'Informatique [Alger]
2 INSA Lyon - Institut National des Sciences Appliquées de Lyon
3 LIRIS - Laboratoire d'InfoRmatique en Image et Systèmes d'information
4 DRIM - Distribution, Recherche d'Information et Mobilité

Preserving the privacy of mobility data has been the center of active research in the last decade as this data may reveal sensitive information about individuals (e.g., home, work places, political, religious, sexual preferences). In this context, a large variety of Location Privacy Protection Mechanisms (LPPMs) have been proposed. To employ LPPMs more effectively for the benefits of the users' privacy, adaptive solutions that dynamically combine LPPMs have also been investigated. These solutions apply various LPPMs on a given trace and choose the one that better meets privacy and utility requirements. To meet this objective, adaptive solutions often rely on a trusted proxy that gathers users' traces and apply LPPMs locally. In this paper we release this assumption by designing an approach, i.e., EDEN, where mobility data never leaves the user's device before being protected by an appropriate LPPM. Experimental evaluation performed on real mobility traces shows that EDEN with its local and adaptive strategy outperforms individual LPPMs both in terms of privacy and utility metrics.

Mots clés: Mobility Data; Location Privacy; Re-identification Attacks; Federated learning

Domaines: Informatique [cs]

Yanis Meziani, Besma Khalfoun, Sara Bouchenak, Sonia Ben Mokhtar, Vlad Nitu. Enforcing Adaptive Location Privacy with Federated Learning. COMPAS 2020 : parallémisme, Architecture, Système/ Temps réel, Jun 2020, Lyon, France. ⟨hal-03339363⟩