Secure and Private Deep Learning with Pysyft - Democast 4

Welcome back, friends! This month, we had the pleasure of sitting down with Andrew Trask, PhD Student at the University of Oxford, and leader of the OpenMined community, for our latest installment of TWIML Democast. Some of you might remember Andrew from the podcast, where he joined Sam in episode #241 to discuss Privacy-Preserving Decentralized Data Science, a great precursor for the video you’re about to watch. OpenMined is focused on “making the world more privacy-preserving by lowering the barrier-to-entry to private AI technologies.” Since our initial conversation with Andrew, the OpenMined community has exploded, with now over 7000 members on Slack, and a recently introduced research arm, OpenMined Research.

In this Democast, we focus on PySyft, their Python library for private machine learning. Andrew breaks down some of the core features of PySyft, including:

• The basics and flow of ML in PySyft
• Remote execution
• The role of encryption, specifically “Secure, Multi-Party Computation,” and Homomorphic Encryption
• Federated Learning
• A look at the OpenMined Roadmap, and what’s next for the project.

Below you’ll find the resources mentioned in the conversation, and please feel free to send any feedback on this conversation or the Democast format in a comment below.

• Connect with Andrew on Twitter!
• CrypTen
• Autograd
• Want to learn more about Federated Learning?
• Maximizing Privacy and Effectiveness in COVID-19 Apps
• #241 – Privacy-Preserving Decentralized Data Science with Andrew Trask