Smooth binary mechanism

An implementation of the smooth binary mechanism as described in the paper A Smooth Binary Mechanism for Efficient Private Continual Observation.

Files included

This repo consists of two files containing code.

mechanism_noise_generation.py

Contains implementation of the binary mechanism and the smooth binary mechanism (for $\rho$-zCDP). Both are implemented as generators where the $i^{th}$ iterate provides the noise that either mechanism would output at time $t=i$.

example.ipynb

A Jupyter Notebook containing a simple example where the noise generators are used to release, under $\rho$-zCDP, all prefix sums on a randomly chosen binary input stream.

Dependencies

Only numpy is needed for including mechanism_noise_generation.py. matplotlib in (addition to jupyter notebook itself) is needed to run example.ipynb.