DPFreqProf: Differentially-Private Software Frequency Profiling Under Linear Constraints

This directory contains the code for randomization described in the paper. Below are the folders that contain essential code and results:

• common and integer-lp contain the common utility classes and the Matlab code
• le-pairs-csv contains the <= pairs using the static analysis from our CC’20 paper
• laplace contains the Laplace-mechanism-based randomization described in the paper
• randomized-response contains the code for the randomized-response-based randomization
• stats includes code for printing statistics

Prerequisites

• Java 1.8+
• MATLAB: Please follow the instructions on MATLAB’s website for installation and then export matlab to PATH of executables.

Run

Clone the repository:

$ git clone https://github.com/presto-osu/dp-freq-prof.git
$ cd dp-freq-prof

Download and extract dataset.tar.gz to dp-freq-prof and execute the following commands to reproduce the experimental results:

$ cd code
$ bash runall_experiments_comparison.sh # comparison of the two randomizers (Sec. 7.1)
$ bash runall_rr_experiments.sh # randomized response (Sec. 7.2 & 7.3)
$ bash runall_lap_experiments.sh # Laplace mechanism (Sec. 7.2 & 7.3)
$ export PARTITION=500  && bash runall_experiments_comparison.sh # high-similarity (Sec. 7.5)
$ export PARTITION=-500 && bash runall_experiments_comparison.sh # low-similarity (Sec. 7.5)

The log files are stored in folder log.

This site is open source. Improve this page.