Bilkent University
Department of Computer Engineering

---

CS590/690 SEMINAR

 

CURE: Privacy Preserving Machine Learning in the Asymmetrical Computational Resource Setting

 

Yaman Yagiz Tasbag
MS Student
(Supervisor: Asst. Prof. Ercüment Çiçek)
Computer Engineering Department
Bilkent University

CURE: Privacy Preserving Machine Learning in the Asymmetrical Computational Resource Setting Abstract: Training deep neural networks often need large scale datasets, which has to be stored and processed on cloud servers due to computational bottlenecks and stringent privacy regulations in domains like healthcare. One popular distributed model training framework is split learning, where layers of the model is split and distributed among the client and the server. While split learning was claimed to be privacy-preserving as the server does not have access to the full parameter set, the communication of intermediate outputs and gradients have been shown to leak information in plaintext mode. While there are generic FHE-based solutions which can be applied to this setting, the computational burden is prohibitive. We propose CURE, a set of FHE-based solutions that only encrypt the server-side of the model. CURE enables secure split learning while offering optimized packing for communication and parallel

 

DATE: 08 November 2021, Monday @ 16:30 Zoom