Concentrated Differential Privacy for Bandits

arXiv:2309.00557v2 Announce Type: replace-cross
Abstract: Bandits serve as the theoretical foundation of sequential learning and an algorithmic foundation of modern recommender systems. However, recommender systems often rely on user-sensitive data, making privacy a critical concern. This paper contributes to the understanding of Differential Privacy (DP) in bandits with a trusted centralised decision-maker, and especially the implications of ensuring zero Concentrated Differential Privacy (zCDP). First, we formalise and compare different adaptations of DP to bandits, depending on the considered input and …

abstract arxiv centralised cs.cr cs.it cs.lg data decision differential differential privacy foundation maker making math.it math.st modern paper privacy recommender systems serve stat.ml stat.th systems type understanding