Differentially-private intermediate tensors example

[TTitcombe]

Feature Description

Develop an example which applies differential privacy to the tensors output from a local user's model before being sent to the computational sever.

Is your feature request related to a problem?

Non-DP tensors could allow a colluding computational server and data holder (in the case of multiple data holders) to identify a user's data using a model inversion attack

What alternatives have you considered?

None

Additional Context