Homomorphic Encryption for secure training

Introduction

Often we need to keep our training data and trained models private for compliance and as well as to preserve our companies secrets.

For example in the field of healthcare we dont want our Cloud GPU providers to have our patients health care data stored cleartext in the server ram. A crafty system administrator could always copy the data for themselves or leak it publically.

Train on Encrypted Data

With Homomorphic Encryption we not only keep the training data encrypted at rest, we also train our model on fully encrypted data, creating a fully encrypted training circuit. No sensitive data can leak out to the node provider whos hardware we are using.

Read more about NVIDIA Federated Learning here plus check the GitHub example at the bottom of the article:https://developer.nvidia.com/blog/federated-learning-with-homomorphic-encryption/

Conclusion

Take advantage of the low cost of GPUS on GPUX to train your neural network while meeting your compliance guidelines.

Get started on GPUX today!