Readme

This repository contains our code for a distributed syntactic privacy protocol presented by Mohammed et al. at TKDD2010 (Mohammed et al. Centralized and Distributed Anonymization for High-Dimensional Healthcare Data. https://doi.org/10.1145/1857947.1857950) as well as the code for our improved protocol variant which addresses weaknesses in the original protocol. The improved variant makes use of the SMPC framework MOTION (https://github.com/encryptogroup/MOTION).

Running the protocol requires building a MOTION-based library. Necessary steps are detailed below.

Build MOTION pandapython lib

• Checkout the repository https://github.com/tompetersen/MOTION and switch to branch panda.
• Follow MOTION build steps detailed in their README.
• Call cmake with -DMOTION_BUILD_EXE=On: $ cmake .. -DMOTION_BUILD_EXE=On
• If successful, build/lib should contain the library pandapython.*.so.

Create environment variable for lib path

• Create an environment variable for this path: $ export MOTION_PANDA_LIB_PATH=/path/to/motion/build/lib/

Setup venv

• Setup your python venv and install requirements
  • $ python3 -m venv venv
  • $ . venv/bin/activate
  • $ pip install -r ours/requirements.txt

Evaluation

• To compare the performance of our vs. the original protocol
  • $ pip install -r eval/requirements.txt
  • $ ./run.sh
  • This creates eval files in eval/data/...
  • $ python eval.py creates figures in img and outputs LaTeX table content
• To show the influence of various QID combinations
  • $./run_arb_qid.sh creates eval files in eval/data/...
  • $ python eval_arb_qid.sh creates figures
• run.sh is also a good starting point for running the protocol yourself manually

Test datasets

• There are two datasets used in the evaluation currently
  • ADULT: The famous Census Income dataset (https://doi.org/10.24432/C5XW20)
  • Medical: A synthetically generated dataset containing some QIDs and additional medical markers.