Privacy Attacks in Decentralized Learning

arXiv:2402.10001v1 Announce Type: new
Abstract: Decentralized Gradient Descent (D-GD) allows a set of users to perform collaborative learning without sharing their data by iteratively averaging local model updates with their neighbors in a network graph. The absence of direct communication between non-neighbor nodes might lead to the belief that users cannot infer precise information about the data of others. In this work, we demonstrate the opposite, by proposing the first attack against D-GD that enables a user (or set of …

abstract arxiv attacks belief collaborative communication cs.cr cs.lg data decentralized gradient graph information neighbors network nodes privacy set type updates