computationalprivacy / querysnout

Source code of ACM CCS 2022 paper "QuerySnout: Automating the Discovery of Attribute Inference Attacks against Query-Based Systems" by Ana-Maria Cretu, Florimond Houssiau, Antoine Cully, and Yves-Alexandre de Montjoye

QuerySnout 🐽: Automating the Discovery of Attribute Inference Attacks against Query-Based Systems

This is the code repository for the "QuerySnout: Automating the Discovery of Attribute Inference Attacks against Query-Based Systems" paper presented at the ACM CCS 2022 conference https://dl.acm.org/doi/abs/10.1145/3548606.3560581.

1 - Environment

To reproduce the paper experiments, please install the conda environment from the .yml file as specified below.

conda env create -f apa.yml

2 - Dataset preprocessing

Run the notebooks/dataset_discretization.ipynb notebook; the discretized dataset will be saved automatically under the datasets folder.

3 - Installing the QBS software

QuerySnout uses an optimized implementation of query-based systems, implemented in a C module interfaced with Python (the cqbs module). To build and install this package, go to the src/optimized_qbs directory and run the following command:

python setup.py install

Note that this repository does not contain an implementation of Diffix for intellectual property reasons. Using the Diffix QBS in the library will produce a warning and return unnoised results. We provide instructions in the code on how to adapt it to add noise similarly to Diffix (or other QBSes).

4 - Running the attack (with several repetitions)

For the AUXILIARY and EXACT-BUT-1 scenarios, run each script by replacing DATASET with "census", "adults" and "insurance".

./automated_attack_table_builder.sh DATASET NUM_PROCS

./automated_attack_dp_laplace.sh DATASET EPSILON NUM_PROCS

./automated_attack_qbs_simple.sh DATASET THRESHOLD SIGMA NUM_PROCS

Note: to run the attack against Diffix, you need to first update the C QBS code as described in src/optimized_qbs/qbs.c. You may then run the script:

./automated_attack_diffix.sh DATASET NUM_PROCS

How to cite

Ana-Maria Creţu, Florimond Houssiau, Antoine Cully, and Yves-Alexandre de Montjoye. 2022. QuerySnout: Automating the Discovery of Attribute Inference Attacks against Query-Based Systems. In Proceedings of the 2022 ACM SIGSAC Conference on Computer and Communications Security (CCS ’22), November 7–11, 2022, Los Angeles, CA, USA. ACM, New York, NY, USA, 24 pages. https://doi.org/10.1145/3548606.3560581