AI-SDC / SACRO-ML

Collection of tools and resources for managing the statistical disclosure control of trained machine learning models

https://ai-sdc.github.io/SACRO-ML/

AI-SDC/SACRO-ML Issues

Remove preprocessing data loaders?
Updated 2 months ago
Improve/extend CLI prompt tools
Updated 2 months ago
new version of attacks/target.py should load more than just pkl files
Updated 2 months ago
need to query MIA risk for different input groups e.g. gender etc
Updated 2 months ago
add src code diagram to contrib.md or wiki if more appropriate
Updated 3 months ago
clean up repository and update packaging
Closed 3 months ago
structural attacks are nor included in documentation
Closed 3 months ago
Add structural attack support for ANNs
Closed 3 months ago
Need a simple document for researchers & TRE staff to agree what files researcher will provide (and why)
Updated 3 months ago
Clean up code docstrings and type hints
Updated 4 months ago
Restore 100% code coverage
Closed 2 months ago1
Fix code coverage reporting
Closed 4 months ago
Remove useless pylint suppression
Closed 4 months ago
ROC Plot axis incorrectly labelled
Closed 4 months ago
Can the uuid package be removed from dependencies?
Closed 4 months ago1
Dependencies
Closed 4 months ago3
Need to increase the visibility of the user stories
Closed 2 months ago2
structural attack for adaboost accessing wrong param name
Closed 10 months ago
Add support for PyTorch
Updated 10 months ago
Structural attack fails in unnecessary disclosure. with xgboost if min_child_weight is not specified
Closed 3 months ago
Check formulation of attribute inference metric
Updated a month ago
report.py: error removing a file that does not exist
Closed a year ago
Example Notebooks should not change user paths but should run using installed acro
Closed 10 months ago
safemodel should not enforce apply_constaints=True
Updated a year ago
Likelihood attack reproducibility
Closed a month ago6
Add support for RUNS metric
Updated a year ago1
Create sub-class 'StructuralAttack' for traditional (and other) risk measures of target modules
Closed 10 months ago1
Automatically format docstrings for consistency
Closed a year ago
Output Directory and JSON filename Issue
Closed a year ago
Single entry point for running multiple attacks
Closed a year ago
Enabling Attribute Inference Attack to Run From Command Line
Closed a year ago
Copy generated files inside attack_report_formatter
Closed a year ago
examples/safemodel_attack_integration_bothcalls.py has to be run inside a 'if __name__ == '__main__'' block
Closed a year ago6
Worst case attack fails when n_dummy_reps = 0
Closed a year ago
Update dependencies for tensorflow-privacy 0.8.10
Closed a year ago4
Add the functionality to use configuration and target files
Closed a year ago
Attribute Attack should report confidence that training set is not more vulnerable than test
Updated a year ago
Invalid command line argument in worst case attack example
Closed a year ago
lira config file pipeline doesn't allow setting n_shadow_models
Closed a year ago
Enhance tests/ request_release test
Closed a year ago
Add all model hyperparameters to safemodel output json
Closed a year ago
Create SafeXGBoostClassifier class
Updated a year ago
Subsuming the attack args classes within the attack classes
Closed a year ago4
Streamlining the outputs produced
Closed a year ago3
OS system calls rely on calling "python"
Closed a year ago
Creation of JSON attack summary should come from a single file
Closed a year ago
SAFEmodels should report generalisation errors
Closed a year ago2
Refactor lira attack test to use fixtures and avoid repetition
Closed a year ago
Standard config file / parameter input
Closed a year ago2
Bringing AIA into the same output format
Closed a year ago