Added support for BEYOND [ICML-2024] #2489

allenhzy · 2024-09-03T08:45:47Z

Description

This pull request adds the support of the BEYOND Detection method proposed in [1].

[1] Be Your Own Neighborhood: Detecting Adversarial Example by the Neighborhood Relations Built on Self-Supervised Learning. ICML. 2024[Paper]

Type of change

Please check all relevant options.

Improvement (non-breaking)
Bug fix (non-breaking)
New feature (non-breaking)
Breaking change (fix or feature that would cause existing functionality to not work as expected)
This change requires a documentation update

Testing

Please describe the tests that you ran to verify your changes. Consider listing any relevant details of your test configuration.

Unit Test

Test Configuration:

OS: Ubuntu 18.04
Python version: 3.9.19
ART version or commit number: 1.18.1
Pytorch: 2.4.0
cudnn version: 90100

Checklist

My code follows the style guidelines of this project
I have performed a self-review of my own code
I have commented my code
I have made corresponding changes to the documentation
My changes generate no new warnings
I have added tests that prove my fix is effective or that my feature works
New and existing unit tests pass locally with my changes
My changes have been tested using both CPU and GPU devices

art/defences/detector/evasion/beyond_detector.py

+"""
+from __future__ import absolute_import, division, print_function, unicode_literals, annotations
+
+import abc


art/defences/detector/evasion/beyond_detector.py

+from __future__ import absolute_import, division, print_function, unicode_literals, annotations
+
+import abc
+from typing import Any


art/defences/detector/evasion/beyond_detector.py

+ :param batch_size: Batch size for processing
+ :param nb_epochs: Number of training epochs (not used in this method)
+ """
+ clean_similarities = self._get_metrics(x, batch_size)


tests/defences/detector/evasion/test_beyond_detector.py

+import pytest
+import numpy as np
+
+import sys


tests/defences/detector/evasion/test_beyond_detector.py

+import numpy as np
+
+import sys
+import os 


tests/defences/detector/evasion/test_beyond_detector.py

+import os 
+
+from art.attacks.evasion.fast_gradient import FastGradientMethod
+from art.estimators.classification import PyTorchClassifier


tests/defences/detector/evasion/test_beyond_detector.py

+from art.attacks.evasion.fast_gradient import FastGradientMethod
+from art.estimators.classification import PyTorchClassifier
+from art.defences.detector.evasion import BeyondDetector
+from art.utils import load_dataset, get_file


tests/defences/detector/evasion/test_beyond_detector.py

+def test_beyond_detector(art_warning, get_cifar10, get_ssl_model):
+ try:
+ # Load CIFAR10 data
+ (x_train, y_train), (x_test, y_test), min_, max_ = get_cifar10


tests/defences/detector/evasion/test_beyond_detector.py

+def test_beyond_detector(art_warning, get_cifar10, get_ssl_model):
+ try:
+ # Load CIFAR10 data
+ (x_train, y_train), (x_test, y_test), min_, max_ = get_cifar10


tests/defences/detector/evasion/test_beyond_detector.py

+ detector = BeyondDetector(
+ target_model=target_model,
+ ssl_model=ssl_model,
+ img_augmentation=img_augmentations,
+ aug_num=50,
+ alpha=0.8,
+ K=20,
+ percentile=5
+ )


codecov · 2024-09-04T08:56:28Z

Codecov Report

Attention: Patch coverage is 21.53846% with 51 lines in your changes missing coverage. Please review.

Project coverage is 77.12%. Comparing base (42c6da4) to head (f5ec52c).
Report is 9 commits behind head on dev_1.19.0.

Files with missing lines	Patch %	Lines
art/defences/detector/evasion/beyond_detector.py	20.31%	50 Missing and 1 partial ⚠️

❗ There is a different number of reports uploaded between BASE (42c6da4) and HEAD (f5ec52c). Click for more details.

HEAD has 3 uploads less than BASE

Flag BASE (42c6da4) HEAD (f5ec52c)

27 24

Additional details and impacted files

@@              Coverage Diff               @@
##           dev_1.19.0    #2489      +/-   ##
==============================================
- Coverage       85.23%   77.12%   -8.11%     
==============================================
  Files             329      330       +1     
  Lines           30143    30208      +65     
  Branches         5173     5177       +4     
==============================================
- Hits            25693    23299    -2394     
- Misses           3021     5606    +2585     
+ Partials         1429     1303     -126

Files with missing lines	Coverage Δ
art/defences/detector/evasion/__init__.py	`100.00% <100.00%> (ø)`
art/defences/detector/evasion/beyond_detector.py	`20.31% <20.31%> (ø)`

... and 49 files with indirect coverage changes

beat-buesser

Hi @allenhzy Thank you very much for your pull request! Could you please take a look at my review comments and add the proposed updates?

beat-buesser · 2024-09-19T22:03:21Z

art/defences/detector/evasion/beyond_detector.py

@@ -0,0 +1,163 @@
+# MIT License
+#
+# Copyright (C) The Adversarial Robustness Toolbox (ART) Authors 2023


Suggested change

# Copyright (C) The Adversarial Robustness Toolbox (ART) Authors 2023

# Copyright (C) The Adversarial Robustness Toolbox (ART) Authors 2024

beat-buesser · 2024-09-19T22:03:53Z

art/defences/detector/evasion/beyond_detector.py

+# TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+# SOFTWARE.
+"""
+This module implements the abstract base class for all evasion detectors.


Suggested change

This module implements the abstract base class for all evasion detectors.

This module implements the BEYOND detector for adversarial examples detection.

| Paper link: https://openreview.net/pdf?id=S4LqI6CcJ3

beat-buesser · 2024-09-19T22:05:58Z

art/defences/detector/evasion/beyond_detector.py

+ """
+ BEYOND detector for adversarial samples detection.
+ This detector uses a combination of SSL and target model predictions to detect adversarial samples.
+ """


Suggested change

"""

BEYOND detector for adversarial samples detection.

This detector uses a combination of SSL and target model predictions to detect adversarial samples.

"""

"""

BEYOND detector for adversarial samples detection.

This detector uses a combination of SSL and target model predictions to detect adversarial examples.

| Paper link: https://openreview.net/pdf?id=S4LqI6CcJ3

"""

beat-buesser · 2024-09-19T22:06:24Z

art/defences/detector/evasion/beyond_detector.py

+from __future__ import absolute_import, division, print_function, unicode_literals, annotations
+
+import abc
+from typing import Any


Suggested change

from typing import Any

beat-buesser · 2024-09-19T22:06:33Z

art/defences/detector/evasion/beyond_detector.py

+from __future__ import absolute_import, division, print_function, unicode_literals, annotations
+
+import abc
+from typing import Any


Suggested change

from typing import Any

beat-buesser · 2024-09-19T22:15:14Z

tests/defences/detector/evasion/test_beyond_detector.py

+ (x_train, y_train), (x_test, y_test), min_, max_ = get_cifar10
+
+ # Load models
+ # Download pretrained weights from https://drive.google.com/drive/folders/1ieEdd7hOj2CIl1FQfu4-3RGZmEj-mesi?usp=sharing


How large are the downloaded files? Can we store them in the ART repo?

beat-buesser · 2024-09-19T22:15:34Z

tests/defences/detector/evasion/test_beyond_detector.py

+ print(f"Clean Detection Accuracy: {clean_accuracy:.4f}")
+ print(f"Adversarial Detection Accuracy: {adv_accuracy:.4f}")


Please replace print with logger.

beat-buesser · 2024-09-19T22:16:19Z

tests/defences/detector/evasion/test_beyond_detector.py

+ assert nb_true_positives > 0
+ assert nb_true_negatives > 0


Would it be possible to make these assertions more accurate?

beat-buesser · 2024-09-19T22:57:00Z

tests/defences/detector/evasion/test_beyond_detector.py

+ return {'z1': z1, 'z2': z2, 'p1': p1, 'p2': p2}
+
+@pytest.fixture
+def get_cifar10():


I think we can use load_cifar10() directly and remove get_cifar10()(.

beat-buesser · 2024-09-19T22:59:48Z

tests/defences/detector/evasion/test_beyond_detector.py

+ """
+ Loads CIFAR10 dataset.
+ """
+ (x_train, y_train), (x_test, y_test), min_, max_ = load_cifar10()


Add import for fixture load_cifar10().

BEYOND Detector and test case

f5ec52c

github-advanced-security bot found potential problems Sep 4, 2024

View reviewed changes

beat-buesser changed the base branch from main to dev_1.19.0 September 4, 2024 09:00

beat-buesser self-requested a review September 4, 2024 09:04

beat-buesser self-assigned this Sep 4, 2024

beat-buesser added the enhancement New feature or request label Sep 4, 2024

beat-buesser added this to the ART 1.19.0 milestone Sep 4, 2024

beat-buesser changed the base branch from dev_1.19.0 to main September 4, 2024 09:07

beat-buesser changed the base branch from main to dev_1.19.0 September 4, 2024 09:07

beat-buesser requested changes Sep 19, 2024

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Added support for BEYOND [ICML-2024] #2489

Added support for BEYOND [ICML-2024] #2489

allenhzy commented Sep 3, 2024

codecov bot commented Sep 4, 2024 •

edited

Loading

beat-buesser left a comment

beat-buesser Sep 19, 2024

beat-buesser Sep 19, 2024

beat-buesser Sep 19, 2024

beat-buesser Sep 19, 2024

beat-buesser Sep 19, 2024

beat-buesser Sep 19, 2024

beat-buesser Sep 19, 2024

beat-buesser Sep 19, 2024

beat-buesser Sep 19, 2024

beat-buesser Sep 19, 2024

	# Copyright (C) The Adversarial Robustness Toolbox (ART) Authors 2023
	# Copyright (C) The Adversarial Robustness Toolbox (ART) Authors 2024

-This module implements the abstract base class for all evasion detectors.
+This module implements the BEYOND detector for adversarial examples detection.
+| Paper link: https://openreview.net/pdf?id=S4LqI6CcJ3

		print(f"Clean Detection Accuracy: {clean_accuracy:.4f}")
		print(f"Adversarial Detection Accuracy: {adv_accuracy:.4f}")

Added support for BEYOND [ICML-2024] #2489

Are you sure you want to change the base?

Added support for BEYOND [ICML-2024] #2489

Conversation

allenhzy commented Sep 3, 2024

Description

Type of change

Testing

Checklist

codecov bot commented Sep 4, 2024 • edited Loading

Codecov Report

beat-buesser left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

codecov bot commented Sep 4, 2024 •

edited

Loading