Can this framework use encrypted training data to train network models? Are there any relevant cases to learn from? about tf-encrypted

[zicofish]

@dikjstra You can first check whether this example helps in your case. It uses data that is shared by one party.

Otherwise, you might have to clarify what encrypted training data you mean. Is it something that is already secret shared? We can definitely add more use cases if necessary. But from our experience, data is always initially owned and encrypted (secret shared) by someone.

[dikjstra]

@dikjstra You can first check whether this example helps in your case. It uses data that is shared by one party.

Otherwise, you might have to clarify what encrypted training data you mean. Is it something that is already secret shared? We can definitely add more use cases if necessary. But from our experience, data is always initially owned and encrypted (secret shared) by someone.

I see, there is one more question, is there a way for me to view the data transmitted from the client on the server side? Is it PondPrivateTensor?

[zicofish]

@dikjstra It is PondPrivateTensor if you use the Pond protocol. Internally, it represents two pieces of shares on two servers respectively.

[zicofish]

@dikjstra The link I provided includes training, not predictions. Please check the details. And it solves the scenario you describe.

[dikjstra]

@dikjstra The link I provided includes training, not predictions. Please check the details. And it solves the scenario you describe.

What should I do if I want to encrypt and aggregatethe data of multiple clients to the server, and how should I configure the code of the client and server, just like this example

[zicofish]

@dikjstra Check the example at this line. Inside the class TrainingClient, it would load data from a device named training-client and encrypt (secret share) the data onto three servers (server0, server1, server2) that are specified in this file. It would help your understanding if you also take a look at how input is processed in the protocol implementation here.

[dikjstra]

Hello, when I run the "b - Secure Model Serving.ipynb" of this example, the following error occurs：

My environment configuration：pyhton=3.8 tensorflow=2.11 tf-encrypted=0.9.1