Zheng Xu

icon indicating a website link https://scholar.google.com/citations?user=TfWlMTYAAAAJ&hl=en

icon company @google @google-research icon location https://sites.google.com/site/xuzhustc/

Results 2 comments of Zheng Xu

AssertionError: compute_gradients() on the differentially private optimizer was not called.

Hi, Do you use a setting in the error message? Using TFPrivacy optimizer in TF2 Keras and `model.fit` may cause the privacy mechanism to be not effectively applied, probably due...

Federated DPFTRL and Adaptive Clipping noise injection.

Thanks for your interest. The noise multiplier is slightly inflated when using adaptive clipping as suggested in the theory. This can be done once similar to https://github.com/tensorflow/federated/blob/main/tensorflow_federated/python/aggregators/differential_privacy.py#L400