Paper accepted to DSML 2024

TrustDDL: A Privacy-Preserving Byzantine-Robust Distributed Deep Learning Framework


Authors: René Klaus Nikiel, Meghdad Mirabi,Carsten Binnig

We are happy to announce that our paper 'TrustDDL: A Privacy-Preserving Byzantine-Robust Distributed Deep Learning Framework' got accepted to the DSML workshop 2024, a joint workshop with the IEEE DSN 2024 conference.

This paper presents TrustDDL, a distributed deep learning framework designed to address privacy and Byzantine robustness concerns throughout the training and inference phases by integrating additive secret-sharing protocols, a commitment phase, and redundant computation to detect and mitigate Byzantine parties, ensuring uninterrupted protocol execution and reliable output delivery, supported by a security analysis demonstrating its effectiveness against various adversaries, and highlighting its practicality compared to existing distributed machine learning frameworks.

DSN 2024:

DSML 2024 Workshop: