Tricks for Accelerating (encrypted) Prediction As a Service

This page provides codes for reproducing our results. We use two main libraries here:

  • The very awesome SHEEP library prepared by Oliver and Nick to run boolean circuits on ecnrypted data using various homomorphic encryption schemes for our networks.
  • Our own general purpose library, matSHEEP, which provides a programmatic interface to design and visualize logic circuits for deep neural networks in python. Find more abour matSHEEP at this link.

There are two main types of experiments.

For more information, refer to our paper on arxiv.