jvmncs
jvmncs
According to [BadNets](https://arxiv.org/abs/1708.06733), pretrained models can be poisoned. One way to partially mitigate this is to provide a hash for the downloaded files. It's also just good security practice, especially...
Related #474 Right now many of the docstrings required for linting are simple descriptors, but they don't follow the Google style guide. We should fix this for all non-trivial functions....
Continuation of #572 See https://github.com/tf-encrypted/tf-encrypted/pull/572#issuecomment-505219938 for details.
Related #416, #388 We'd like to refactor our layers module to match the keras api spec. I'll work on this once the style guide PR has been submitted (#424)
We are quickly going to be interfacing with quite a few libraries (PySyft, tf-big, scale mamba, ...). We will need a strong suite of integration tests that run with CI....
we might consider using pytest's parametrize functionality instead of the `unittest.TestCase` class where appropriate. it's somewhat of an aesthetic detail, but it might make it easier for new contributors to...
Related to #499 -- see note in `PondPrivatePlaceholder.feed` [here](https://github.com/tf-encrypted/tf-encrypted/blob/c5011d3c44183e8fd8a8728879b3bc37b9e0f90d/tf_encrypted/protocol/pond/pond.py#L1919)
We've occasionally observed non-deterministic tests in some cases. Testing should be done deterministically, so we should track down which tests are non-deterministic and fix them with proper seeding.
After refactor: `tfe.Variable(dtype=backing_tensor, shape=[4, 5], initializer=tf.keras.initializers.zeros)`
From this comment thread: https://github.com/mortendahl/tf-encrypted/pull/352#discussion_r241010696 We'll probably want to be able to convert graphs where some Placeholders represent private inputs and some represent public ones. We may want to wait...