ML Confidential: Machine Learning on Encrypted Data

Kristin Lauter (Microsoft Research, USA)

 

Abstract

This talk will cover practical applications of Homomorphic Encryption including machine learning and genomic computation. The possibility of outsourcing computation to the cloud offers businesses and individuals substantial cost-savings, flexibility, and availability of compute resources, but potentially sacrifices privacy. Homomorphic encryption can help address this problem by allowing the user to upload encrypted data to the cloud, which the cloud can then operate on without having the secret key. The cloud can return encrypted outputs of computations to the user without ever decrypting the data, thus providing hosting of data and services without compromising privacy. Important applications include electronic medical data and financial applications, as well as private targeted advertising. The catch is the degradation of performance and issues of scalability and flexibility. This talk will survey the trade-offs when using homomorphic encryption, and highlight scenarios and functionality where homomorphic encryption seems to be the most appropriate solution. In recent work, we showed that homomorphic encryption can even be used to enable private versions of some basic machine learning algorithms and some genomic computations.