Can concrete-ml Implement Backpropagation for Training Neural Networks with TFHE Encryption?

Can concrete-ml be used to implement and perform backpropagation for training neural networks with TFHE? I’m looking to train neural networks using TFHE and would like to know if concrete-ml supports this functionality.

Hi,

Thank you for your question.

Concrete ML currently only supports FHE inference. However, training in FHE is in our roadmap, but for now, it has to be on clear data.

If you’d like to perform FHE inference using neural networks, you can check out our use-case example.

For more information on Concrete ML limitations, you can visit this link: What is Concrete ML? - Concrete ML.

Thanks !