train a neural network on encrypted values using MPC

[adityapribadi3]

Question

is it possible to train a neural network on encrypted values using Secure Multi-Party Computation and Autograd?

Further Information

I found a blog from pysyft that can perform encrypted train in MNIST Model using MPC, I'm a bit confuse how to use SyMPC in pysyft 0.5 to perform the similar task. The blog can be found here https://blog.openmined.org/encrypted-deep-learning-classification-with-pysyft/

if there any example or any tutorial how to run similarly using SyMPC?

Screenshots

If applicable, add screenshots to help explain your question.

System Information

• OS: MAC OS

Additional Context

Add any other context about the problem here.

[letv3]

Hi all, SMPC topic seems very exiting topic to me, I would be glad to get some update on this old tutorial SyMPC on MNIST as well. Cheers :). But actually I can work on updating this tutorial using up-to-date examples in main SyMPC repo https://github.com/OpenMined/SyMPC/tree/main/examples

[yeungbo]

As the example code shows, I think it is Plaintext Training and encrypted prediction, NOT train a neural network on encrypted values, right?

[adityapribadi3]

yes that's right

[yeungbo]

So, does this issue have any solution idea?

[rasswanth-s]

The secure training notebook example is being worked, @kamathhrishi ping. We have notebook examples for secure inference https://github.com/OpenMined/SyMPC/tree/main/examples as of now

[yeungbo]

The secure training notebook example is being worked, @kamathhrishi ping. We have notebook examples for secure inference https://github.com/OpenMined/SyMPC/tree/main/examples as of now

Thank you Rasswanth, the secure inference does work, but why not secure train? I think the issue here is for the later :)