Code for Learning-augmented private algorithms for multiple quantile release, to appear in ICML 2023.
The scripts static.py, pubpri.py, and online.py are for the experiments in Sections 3.3, 5.1, and 5.2, respectively.
Note some scripts may download potentially large datasets, and the file citibike/worldnews/comments.pkl.zip needs to be unzipped before running online.py.
The Dockerfile describes the Python environment used.
@inproceedings{khodak2023learning,
author={Mikhail Khodak and Kareem Amin and Travis Dick and Sergei Vassilvitskii},
title={Learning-Augmented Private Algorithms for Multiple Quantile Release},
booktitle={Proceedings of the 40th International Conference on Machine Learning},
year={2023}
}