We apply TF Encrypted to a typical deep learning example, providing a good starting point for anyone wishing to get into this rapidly growing field. As shown, using state-of-the-art secure computation techniques to serve predictions on encrypted data requires nothing more than a basic familiarity with deep learning and TensorFlow.
Using TensorFlow as a distributed computation framework for dataflow programs we give a full implementation of a secure computation protocol with networking, in turn enabling optimised machine learning on encrypted data.
We take a typical CNN deep learning model and go through a series of steps that enable both training and prediction to instead be done on encrypted data using the SPDZ protocol.
We have previously seen that redundancy in secret sharing can be used to recover from lost shares. In this third part of the series we use Reed-Solomon decoding methods to see that it can also be used to detect when some shares have been manipulated.
Overview of work done at Snips on applying privacy-enhancing technologies as a start-up building privacy-aware machine learning systems for mobile devices. Mainly centered around secure aggregation for federated learning from user data but also some discussion around privacy from a broader perspective.
Efficient secret sharing requires fast polynomial evaluation and interpolation. In the second part of the series we go through how the well-known Fast Fourier Transform can be used for this.
First part in a series where we look at secret sharing schemes, including the lesser known packed variant of Shamir's scheme, and give full and efficient implementations. We start in this post by looking at the more typical textbook approaches.
We build a simple secure computation protocol from scratch and use it to train simple neural networks for basic boolean functions.