Fast Homomorphic Evaluation of Deep Discretized Neural Networks

Explore a conference talk on fast homomorphic evaluation of deep discretized neural networks, presented at Crypto 2018. Delve into the research by Florian Bourse, Michele Minelli, Matthias Minihold, and Pascal Paillier, which addresses challenges in neural network evaluation using homomorphic encryption. Examine the problem statement, neural network applications in digit recognition, and the current state of the art. Learn about the Multisim approach, discretized network evaluation, and activation issues. Understand the concept of dynamic message space and the overall process. Review experimental results, the proposed framework, and discuss open problems in this field of cryptography and machine learning.

Introduction
Problem Statement
Neural Networks
Digit Recognition
State of the Art
Multisim
Discretized Network
Evaluation
Activation
Issues
Dynamic Message Space
Process Overview
Experimental Results
Framework
Open Problems