Feat: Add three new Membership Inference attacks

[iamsh4shank]

Description

Added three new types of membership inference attacks, these are as follows -

• SIF Attack
• Blind Membership inference Attack
• White Box based Membership Inference Attacks

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.

• I ran test files to test the changes

Test Configuration:

• Ubuntu 20
• 3.9
• 1.15
• TensorFlow / Keras / PyTorch / MXNet version - all in the newest version

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

[iamsh4shank]

@beat-buesser could you review my pull request and let me know if I have to make changes? I am new to the community and would like to add some more Confidentiality attacks.

[beat-buesser]

Hi @iamsh4shank Thank you very much for your interest in ART and working on contributing new inference attacks! We will review this PR as soon as possible.

[codecov-commenter]

Codecov Report

Merging #2271 (f710043) into dev_1.17.0 (c5c6012) will decrease coverage by 7.16%.
Report is 328 commits behind head on dev_1.17.0.
The diff coverage is 14.84%.

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

@@              Coverage Diff               @@
##           dev_1.17.0    #2271      +/-   ##
==============================================
- Coverage       85.62%   78.47%   -7.16%     
==============================================
  Files             308      314       +6     
  Lines           27448    28399     +951     
  Branches         5044     5235     +191     
==============================================
- Hits            23502    22285    -1217     
- Misses           2667     4867    +2200     
+ Partials         1279     1247      -32     

Files	Coverage Δ
...attacks/inference/membership_inference/__init__.py	100.00% <100.00%> (ø)
art/estimators/classification/pytorch.py	86.03% <66.66%> (-0.27%)	⬇️
...rt/attacks/inference/membership_inference/utils.py	23.07% <23.07%> (ø)
...s/inference/membership_inference/blindMI_attack.py	29.62% <29.62%> (ø)
...ership_inference/self_influence_function_attack.py	14.38% <14.38%> (ø)
...ttacks/inference/membership_inference/utils_sif.py	18.30% <18.30%> (ø)
...erence/membership_inference/influence_functions.py	12.40% <12.40%> (ø)
...ttacks/inference/membership_inference/white_box.py	8.26% <8.26%> (ø)

... and 25 files with indirect coverage changes

[github-advanced-security]

CodeQL found more than 10 potential problems in the proposed changes. Check the Files changed tab for more details.

[beat-buesser]

Hi @iamsh4shank It looks like something failed in the new unit tests like unexpected arguments, etc. Could you please take a look?

[iamsh4shank]

Hi @iamsh4shank It looks like something failed in the new unit tests like unexpected arguments, etc. Could you please take a look?

Yep sure will check and update the pull request soon

[iamsh4shank]

@beat-buesser could you please once run the CI/CD?

[iamsh4shank]

@beat-buesser I modified the commit, now it should fix almost all of the checks. I have a question here in this https://github.com/Trusted-AI/adversarial-robustness-toolbox/actions/runs/6130115176/job/16713558817?pr=2271, as here there is no installation for the cv2 that causes a few failures. So should I add cv2 in these files?

[iamsh4shank]

Hi is there any update on this?

[beat-buesser]

Hi @iamsh4shank Sorry for the delay. We think the attacks are a great addition to ART's current inference attacks. We'll add a more detailed technical review soon. A few items I have noticed from a high-level review: could you please start adding unit tests for each attack and function in tests/ following the pytest pattern, update/add all docstrings following ART's format and add typing to all functions and methods?

[iamsh4shank]

Hi @iamsh4shank Sorry for the delay. We think the attacks are a great addition to ART's current inference attacks. We'll add a more detailed technical review soon. A few items I have noticed from a high-level review: could you please start adding unit tests for each attack and function in tests/ following the pytest pattern, update/add all docstrings following ART's format and add typing to all functions and methods?

Yep sure, I will start working on the tests