Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add unit tests for PGD implementation in TF2 #1038

Open
AlexeyKurakin opened this issue Apr 22, 2019 · 7 comments
Open

Add unit tests for PGD implementation in TF2 #1038

AlexeyKurakin opened this issue Apr 22, 2019 · 7 comments

Comments

@AlexeyKurakin
Copy link
Collaborator

No description provided.

@AlexeyKurakin AlexeyKurakin added this to the v4 initial release milestone Apr 22, 2019
@WindQAQ
Copy link
Contributor

WindQAQ commented Jul 2, 2019

Hi @AlexeyKurakin, is anyone working on this? If not, I would like to do this. Thanks!

@npapernot
Copy link
Member

No one is working on TF2 tests, so feel free to take it! Thanks! You can look at #1045 for some guidance.

@WindQAQ
Copy link
Contributor

WindQAQ commented Jul 2, 2019

Sure! Want to make sure something here:

  1. do we need to decorate the tests with run_all_in_graph_and_eager_modes to make them compatible with graph mode?
  2. should we inherit the tests from tf.test.TestCase instead of CleverHansTest? It overrides some methods to make unittest more suitable for tensor.
    Thanks!

@npapernot
Copy link
Member

  1. Let me get back to you on this one
  2. Yes that sounds like a good idea if it simplifies your implementation

@WindQAQ
Copy link
Contributor

WindQAQ commented Jul 2, 2019

Thanks you! I would create a PR for fast_gradient_method first to let you see if it is good enough.

@npapernot
Copy link
Member

Regarding #1, we had a discussion with @AlexeyKurakin and our current stand is that we should aim to verify the possibility of wrapping all attack functions in @tf.function in TF2 but that we do not need to verify compatibility with TF1 graph mode.

@WindQAQ
Copy link
Contributor

WindQAQ commented Jul 3, 2019

Regarding #1, we had a discussion with @AlexeyKurakin and our current stand is that we should aim to verify the possibility of wrapping all attack functions in @tf.function in TF2 but that we do not need to verify compatibility with TF1 graph mode.

Okay! I'm also fine with this.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants