slytHErin: an agile framework for encrypted deep neural network inference

The slytHErin framework, through a hybrid design of homomorphic encryption (HE) and its multi-party variant (MHE), supports privacy-preserving prediction-as-a-service (PaaS) with encryption of the data, the model, or both simultaneously. The framework can flexibly pack inputs, process batches of an arbitrary number of samples, and includes computation optimizations independent of model and settings. The framework is implemented in Go, and a few lines of code suffice to perform encrypted inference on custom deep learning models containing fully connected, convolutional and pooling layers.