mirror of
https://github.com/IBM/ai-privacy-toolkit.git
synced 2026-04-26 05:16:22 +02:00
13 lines
725 B
Python
13 lines
725 B
Python
|
"""
|
||
|
|
Module providing privacy risk assessment for synthetic data.
|
||
|
|
|
||
|
|
The main interface, ``DatasetAttack``, with the ``assess_privacy()`` main method assumes the availability of the
|
||
|
|
training data, holdout data and synthetic data at the time of the privacy evaluation.
|
||
|
|
It is to be implemented by concrete assessment methods, which can run the assessment on a per-record level,
|
||
|
|
or on the whole dataset.
|
||
|
|
The abstract class ``DatasetAttackMembership`` implements the ``DatasetAttack`` interface, but adds the result
|
||
|
|
of the membership inference attack, so that the final score contains both the membership inference attack result
|
||
|
|
for further analysis and the calculated score.
|
||
|
|
"""
|
||
|
|
from apt.risk.data_assessment import dataset_attack
|