TenSEAL: A Library for Encrypted Tensor Operations Using Homomorphic Encryption

Machine learning algorithms have achieved remarkable results and are widely in a variety of domains. These algorithms often rely on sensitive and data such as medical and financial records. Therefore, it is vital to further attention regarding privacy threats and corresponding defensive applied to machine learning models. In this paper, we present, an open-source library for Privacy-Preserving Machine Learning using Encryption that can be easily integrated within popular machine frameworks. We benchmark our implementation using MNIST and show that encrypted convolutional neural network can be evaluated in less than a, using less than half a megabyte of communication.