Efficient Evaluation of Activation Functions over Encrypted Data

Explore a 23-minute conference talk presented at the 2nd Deep Learning and Security Workshop during the 2019 IEEE Symposium on Security & Privacy. Delve into a novel method for approximating bounded activation functions with encrypted input data, demonstrated through simulations in machine learning tasks. Learn about the application of this technique in a Variational Autoencoder for learning latent representations of MNIST data and an MNIST image classifier. Gain insights into the intersection of cryptography and deep learning, and understand how this approach can enhance privacy and security in machine learning applications.