Deep Learning Example- Data Encryption

Hello everyone,

I’m going through this notebook ( )line by line. The only thing I’m not understanding is the specific line of code where we encrypt the data. Which line of the code is to encrypt the input data? Could someone please clarify this for me?

Thank you!

Hello @aidenparker,
The line that are somewhat looking for lies within the test_with_concrete function (cell 6) :

        # Quantize the inputs and cast to appropriate data type
        y_pred = quantized_module.forward(data, fhe=fhe_mode)

Here, foward is a method that groups several major steps all together. Basically, it :

  • quantizes the inputs (in the clear)
  • encrypts the quantized inputs
  • runs the inference in FHE using the encrypted inputs
  • decrypts the (quantized) outputs
  • de-quantizes the ouputs (in the clear)

We have a documentation page that gives an example on how to do these steps separately with built-in models. While the example is about a LogisticRegression model and its predict method, the ideas and APIs behind remain the same for a QuantizedModule and its forward method.

Besides, if you are more interested in having these steps separated for model deployment in a Client-Server setting, we also provide a specific API for that. You will find more information in the cloud inference section, as well as the deployment documentation along its notebook example.

Hope this helps !

1 Like