Oblivious Multi-Party Machine Learning on Trusted Processors

Speaker Organization:  Microsoft Research, Cambridge, UK) Abstract: Privacy-preserving multi-party machine learning allows multipleorganizations to perform collaborative data analytics while guaranteeing the privacy of their individual datasets. Using trusted SGX-processors for this task yields high performance, but requires a careful selection, adaptation, and implementation of machine-learning algorithms to provably prevent the exploitation of any side channels induced by data-dependent access patterns. In this talk, I will present our data-oblivious counterparts of several machine learning algorithms including support vector machines, matrix factorization, neural networks and decision trees. These algorithms aredesigned to access memory without revealing secret information about theirinput. We use algorithmic techniques as well as platform specific hardwarefeatures to ensure that only public information, such as dataset size, is revealed. I will show that our efficient implementation on Intel Skylake processors scales up to large, realistic datasets, with overheads several…


Link to Full Article: Oblivious Multi-Party Machine Learning on Trusted Processors