USENIX Lightning Talk: Fully decentralized privacy-preserving machine learning framework

Abstract

In this talk, we introduce Kaires, a fully decentralized machine learning framework that is private, fair, robust, and auditable. Kaires enables data consumers to build machine learning models while maintaining the privacy of individual data providers as well as the confidentiality of the model parameters. To achieve these goals, Kaires employs a fault-tolerant multi-party computation protocol for computing aggregate statistics backed with a system for auditable management of private data. Besides enabling the data providers to have fine-grained access-control over their data, the system is also fault-tolerant with no single point of compromise or failure.

Date
Location
Santa Clara, CA, USA
Links