Description

This GitHub repository houses an implementation of the Private-GSD mechanism as outlined in the research paper, "Generating Private Synthetic Data with Genetic Algorithms," presented at the 40th International Conference on Machine Learning in 2023.

The Private-GSD mechanism is a specialized synthetic data generation tool, designed to preserve different classes of statistical queries derived from a given dataset while adhering to the principles of differential privacy.

Example

Visit this Colab link to start using Private-GSD.

Setup

Set up conda environment

conda create -n gsd python=3.9

conda activate gsd 
pip install --upgrade pip

Install via setuptools

cd ~/
git clone https://github.com/giusevtr/private_gsd.git
cd ~/private_gsd 
pip install -e .

Install JAX separately. For example,

pip install --upgrade  "jax[cuda11_cudnn82]==0.4.6" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html

Please make sure that the command you execute matches your system (i.e., tpu vs. gpu, right CUDA/cuDNN versions, etc.)

Download and preprocess datasets using dp-data.

cd ~/private_gsd 
git clone https://github.com/terranceliu/dp-data
cd ~/private_gsd/dp-data
pip install -e .
./preprocess_all.sh
cd ~/private_gsd