privacy

privacy copied to clipboard

AttributeError: 'Conv2D' object has no attribute 'apply'

2

I'm not sure if I'm doing something wrong, but I just ran: pip install tensorflow-privacy then: python mnist_dpsgd_tutorial.py And I'm getting the error: File "/privacy/tutorials/mnist_dpsgd_tutorial_common.py", line 23, in get_cnn_model y...

JosephMPedersen

AssertionError: compute_gradients() on the differentially private optimizer was not called. Which means that the training is not differentially private. It happens for example in Keras training in TensorFlow 2.0+.

3

Dear Developers, Thank you for implementing and maintaining such a great repository. I tried to train my model without eager execution in `TensorFlow version 2.3.1`. Some code snippets are below:...

mcps5601

ftramer

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

10

I randomly run into the following error: ```bash AssertionError: compute_gradients() on the differentially private optimizer was not called. ``` The reason is unknown, as the above error does not seem...

mayankshah1607

compute_dp_sgd_privacy_lib.py relies on deprecated rdp_accountant.py

2

Estimating the privacy budget currently doesn't work. The library[1] relies on a deprecated rdp_accountant[2]. Specifically, [3] is missing from the new API. [1]: https://github.com/tensorflow/privacy/blob/master/tensorflow_privacy/privacy/analysis/compute_dp_sgd_privacy_lib.py#L20-L21 [2]: https://github.com/tensorflow/privacy/blob/master/tensorflow_privacy/privacy/analysis/rdp_accountant.py#L15 [3]: https://github.com/tensorflow/privacy/blob/master/tensorflow_privacy/privacy/analysis/rdp_accountant.py#L205

CdavM

Relax package requirements and follow PEP 517

1

Hi there 👋🏼 The latest version of `tensorflow_privacy` (`0.8.0`) seems to have very strict package requirements --> **it's not allowing the installation of things such as `matplotlib==3.4` or `pandas==1.3.5` in...

simonhkswan

This will help reproduce Figure 7 in https://openreview.net/pdf?id=-70L8lpp9DF

npapernot

Extremely high epsilon in MNIST Keras tutorials

1

The `noise_multiplier` in the MNIST Keras tutorials is extremely low (0.1). This results in an epsilon of 66416.55 after 60 epochs of training. Looks like this is a typo introduced...

vdasu

Updated URL for recommended tutorials

1

This old URL 404'd: https://www.tensorflow.org/tutorials/keras/basic_classification Replaced with proper link to tutorial https://www.tensorflow.org/tutorials/keras/classification

ssh-keyz

Government of India

Mahadeb73