Frontiers in Machine Learning: Accelerating Machine Learning with Confidential Computing

In the recent years, Machine Learning (ML) has facilitated key applications, such as medical imaging, video analytics, and financial forecasting. Understanding the massive computing requirements of ML, cloud providers have been investing in accelerated ML computing and a range of ML services. A key concern in such systems, however, is the privacy of the sensitive data being analyzed and the confidentiality of the trained models. Confidential cloud computing provides a vehicle for privacy-preserving ML, enabling multiple entities to collaborate and train accurate models using sensitive data, and to serve these models with assurance that their data and models remain protected, even from privileged attackers. In this session, our speakers will demonstrate applications and advancements in Confidential ML: (i) how confidential computing hardware can accelerate multi-party and collaborative training, creating an incentive for data sharing; and (ii) how emerging cloud accelerator systems can be re-designed to deliver strong privacy guarantees, overcoming the limited performance of CPU-based confidential computing.

Date:
Haut-parleurs:
Antoine Delignat-Lavaud, Raluca Ada Popa, Emmett Witchel
Affiliation:
Microsoft Research, University of California, Berkeley, University of Texas at Austin

Regardez suivant