Output details

ML Confidential: Machine Learning on Encrypted Data.

<18> How can a cloud service perform a machine learning service for a client without the client having to disclose the training/test data to the cloud? This astonishing feat is accomplished by implementing machine learning algorithms such as Fisher’s Linear Discriminant under Somewhat Homomorphic Encryption (SHE), a framework which allows for homomorphic encryption on a limited function class. The technical breakthrough is the implementation of machine learning algorithms using only fixed degree polynomials, which allow for SHE at relatively small computational cost. The work may have important applications in predictive/diagnostic services in medical applications, and in particular genomic analysis.