CrypTFlow - Secure TensorFlow Inference

Explore a groundbreaking system for secure TensorFlow inference in this IEEE conference talk. Dive into CRYPTFLOW, a pioneering solution that automatically converts TensorFlow inference code into Secure Multi-party Computation (MPC) protocols. Learn about the three key components: Athos, an end-to-end compiler; Porthos, an improved semi-honest 3-party protocol; and Aramis, a novel technique for malicious secure MPC protocols. Discover how CRYPTFLOW achieves secure inference of complex neural networks like RESNET50 and DENSENET121 over the ImageNet dataset, outperforming prior work in both semi-honest and malicious security scenarios. Gain insights into the system's architecture, its components, and the innovative approach to malicious security using hardware with integrity guarantees.

Introduction
The Problem
What is MPC
Prior work
Goal
Components
Compilation
Malicious Security
Aramis
Summary