Experiments of the paper "The statistical nature of leakage in SSE schemes and its role in passive attacks"

Authors: Marc Damie, Jean-Benoist Leger, Florian Hahn, and Andreas Peter.

For any question about the code or the paper, contact Marc Damie.

Install

The install process is straightforward: bash setup.sh. This script installs the Python dependencies and download the datasets.

Reproduce

To repoduce our results, you need to run python3 generate_results.py. This script launches all the experiments one by one. The results will be stored in multiple CSV files in a results folder.

To generate the figures, you need to run python3 generate_figures.py. This script generates all figures one by one using the CSV files generated by the previous script.

Repository structure

The folder src/ contains the following elements:

• document_extraction.py contains the functions to process the datasets. Each dataset has a different format so there are dedicated functions for each datasets.
• keyword_extraction.py contains the functions to process the extracted documents. These functions use multiprocessing to extract keywords efficiently.
• simulation_utils.py contains all auxiliary functions necessary to simulate attacks; e.g., adversary knowledge generation.
• attacks/ contains the functions to perform the score and IHOP attacks.

Carbon emission estimation

In line with recent works in ML, we measured the carbon emission related to our experiments using the package CodeCarbon.