Implement Interval Bound Propagation Training #2044
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
GiulioZizzo
force-pushed
the
IBP_Training
branch
from
45f8262
to
d900c5f
Compare
Codecov Report
📣 This organization is not using Codecov’s GitHub App Integration. We recommend you install it so Codecov can continue to function properly for your repositories. Learn more @@ Coverage Diff @@
## dev_1.14.0 #2044 +/- ##
==============================================
+ Coverage 76.94% 85.58% +8.63%
==============================================
Files 291 292 +1
Lines 25612 25798 +186
Branches 4631 4665 +34
==============================================
+ Hits 19708 22079 +2371
+ Misses 4871 2529 -2342
- Partials 1033 1190 +157
|
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
beat-buesser
changed the title
Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
| ABLATOR_TYPE = Union[BlockAblator, ColumnAblator] # pylint: disable=C0103 | ||
|
|
||
| CERTIFIER_TYPE = Union[PytorchDeepZ] # pylint: disable=C0103 | ||
| IBP_CERTIFIER_TYPE = Union[PyTorchIBPClassifier] # pylint: disable=C0103 |
Check notice
Code scanning / CodeQL
Unused global variable
beat-buesser
requested changes
Mar 7, 2023
There was a problem hiding this comment.
Hi @GiulioZizzo Thank you very much for your pull request in Interval Bound Propagation training! I have left a few comments, what do you think?
art/defences/trainer/__init__.py
Outdated
| @@ -4,6 +4,7 @@ | |||
| from art.defences.trainer.trainer import Trainer | |||
| from art.defences.trainer.adversarial_trainer import AdversarialTrainer | |||
| from art.defences.trainer.certified_adversarial_trainer_pytorch import AdversarialTrainerCertifiedPytorch | |||
| from art.defences.trainer.ibp_certified_trainer_pytorch import AdversarialTrainerCertifiedIBPPytorch | |||
There was a problem hiding this comment.
Suggested change
| from art.defences.trainer.ibp_certified_trainer_pytorch import AdversarialTrainerCertifiedIBPPytorch | |
| from art.defences.trainer.ibp_certified_trainer_pytorch import AdversarialTrainerCertifiedIBPPyTorch |
There was a problem hiding this comment.
Agree, name changed!
Comment on lines
+210
to
+215
| This classifier has 3 modes which can be set via: classifier.model.set_forward_mode('mode') | ||
|
|
||
| 'mode' can be one of: | ||
| + 'abstract': When we wish to certifiy datapoints and have abstract predictions | ||
| + 'concrete': When normal predictions need to be made | ||
| + 'attack': When we are interfacing with an ART attack (for example PGD). |
There was a problem hiding this comment.
Does this docstring build correctly?
There was a problem hiding this comment.
checked, all ok!
| @@ -0,0 +1,462 @@ | |||
| # MIT License | |||
There was a problem hiding this comment.
Please update docs and check if new docstrings build correctly.
…with much less protected member access Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
Co-authored-by: Beat Buesser <49047826+beat-buesser@users.noreply.github.com> Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
Co-authored-by: Beat Buesser <49047826+beat-buesser@users.noreply.github.com> Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
Co-authored-by: Beat Buesser <49047826+beat-buesser@users.noreply.github.com> Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
GiulioZizzo
force-pushed
the
IBP_Training
branch
from
887660e
to
879d8fd
Compare
Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
Signed-off-by: GiulioZizzo <giulio.zizzo@yahoo.co.uk>
beat-buesser
approved these changes
Mar 11, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
In this PR we include adversarial training with interval bound propagation. With careful training this can achieve excellent certified robustness.
Fixes #2037
Type of change
Please check all relevant options.
Testing
Tests can be found in test_interval.py. We check that the certification gives the correct results, runs without errors, propagates gradients correctly and when running without bounds gives the same results as regular neural networks.
Test Configuration:
Checklist