This package contains the algorithm and experiments for the following paper:
Differentially Private Learning of Structured Discrete Distributions
Ilias Diakonikolas, Moritz Hardt, Ludwig Schmidt
NIPS 2015
The code is written in Julia (v0.5). The script src/histogram_experiment.jl runs the experiments in the paper for synthetic distributions (mixtures of Gaussian, Beta, and Gamma distributions).
For an example of how to use the code, see the Jupyter notebook examples/histogram_approximation.ipynb.
The three plots above are generated in the notebook examples/website_plots.ipynb.