PyTorch implementation of GeoPointGAN

(GeoPointGAN generated and privatized data: 311 caller locations in New York.)

This is the official repository for the paper GeoPointGAN: Synthetic Spatial Data with Local Label Differential Privacy. GeoPointGAN is a generative model for geographic point coordinates and also includes a privacy mechanism.

(GeoPointGAN pipline, including privacy mechanism.)

Structure

The source code for GeoPointGAN can be accessed in the src folder, our training datasets (those that are not hosted externally) can be found in the data folder and the notebooks folder contains an example notebook.

You can open the interactive example notebook straight away via Google Colab:

Citation

If you want to cite our work, you can use the following reference:

@misc{cunningham2022geopointgan,
    title={GeoPointGAN: Synthetic Spatial Data with Local Label Differential Privacy},
    author={Teddy Cunningham and Konstantin Klemmer and Hongkai Wen and Hakan Ferhatosmanoglu},
    year={2022},
    eprint={2205.08886},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}