Training and Fine-Tuning Examples

These examples train or fine-tune networks with the data, the activations and the parameters encrypted, using the standard forward and backward pass controllers described in Encrypted Training. They are built with -DMODE=TRAINING (HETrainingModels/).

Program

Layout

Purpose

tuneResNet20

Standard

One fine-tuning step of a pretrained ResNet-20 on 16 CIFAR-10 images (multithreaded).

trainModules

Standard

Validation modules: the training primitives on small deterministic inputs.