UnSplit: Data-Oblivious Model Inversion, Model Stealing, and Label Inference Attacks Against Split Learning. (arXiv:2108.09033v2 [cs.CR] UPDATED)

Training deep neural networks often forces users to work in a distributed or
outsourced setting, accompanied with privacy concerns. Split learning aims to
address this concern by distributing the model among a client and a server. The
scheme supposedly provides privacy, since the server cannot see the clients'
models and inputs. We show that this is not true via two novel attacks. (1) We
show that an honest-but-curious split learning server, equipped only with the
knowledge of the client neural …

attacks data oblivious split learning stealing