Automatic Tuning of Privacy Budgets in Input-Discriminative Local Differential Privacy

LDP (Local Differential Privacy) and its variants have been recently studied to analyze personal data collected from IoT (Internet of Things) devices while strongly protecting user privacy. In particular, a recent study proposes general privacy notion called ID-LDP (Input-Discriminative LDP), which introduces budget for each input value deal with different levels sensitivity. However, it is unclear how set an appropriate value, especially in current situations where re-identification considered major risk, e.g., GDPR. Moreover, the possible number values can be very large IoT. Consequently, also extremely difficult manually check whether appropriate. this paper, we propose algorithms automatically tune budgets so that obfuscated prevent re-identification. We new instance OneID-LDP (One-Budget Input-Discriminative LDP) high utility. Through comprehensive experiments using four real datasets, show existing instances lack either utility or – they overprotect are vulnerable attacks. Then our mechanisms tuning algorithm provide much higher than preventing