Evaluation of Google's differential privacy tool, along with the modifications made, on a real world database.
-> A basic, non-technical introduction to PPA and differential privacy: https://desfontain.es/privacy/differential-privacy-awesomeness.html
-> The tool can be found at: https://github.com/google/differential-privacy (Only the modified wrapper is present in the repo, so the tool has to be built before cloning)
-> The differentially privacy paper can be found at: https://arxiv.org/pdf/1909.01917.pdf
-> Tool Requirements:
- Linux OS
- Bazel v.3
- PostgreSQL v.11.0
- Python 3.6 or higher
Note: Fast and stable internet is absolutely necessary in order to build the tool successfully.
-> Modifications made include:
- Conversion of normal SQL queries to the intrinsically private queries that the tool requires
- Improvement in performance overhead using a different mechanism to calculate lower and upper bounds.
-> Database.zip contains the SQL files to create the testing database.
- To run a normal query:
python3 Privacy.py <database name> "<Query without the ';' at the end>" - To run the test script (only works for the UBER database:
python3 test_privacy.py "<test input file that has all the queries>" - To run the modified WWM file:
python3 privacy_wwm.py <database name> "<Query without the ';' at the end>" - To run the string matching modified file:
python3 str_privacy.py <database name> "<Query without the ';' at the end>"
- Adithi Satish
- Avnish Goel
- Manah Shetty