Sharp Composition Bounds for Gaussian Differential Privacy via Edgeworth Expansion

Authors: Qinqing Zheng, Jinshuo Dong, Qi Long, Weijie Su

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we present numerical experiments to compare the Edgeworth approximation and the CLT approximation.
Researcher Affiliation Academia 1University of Pennsylvania.
Pseudocode No The paper does not contain any pseudocode or algorithm blocks.
Open Source Code Yes The code is available at https://github.com/ enosair/gdp-edgeworth.
Open Datasets No The experiments are numerical comparisons of theoretical bounds for differential privacy, not experiments on a real-world dataset with traditional training/test/validation splits. Therefore, no training dataset is explicitly used or made available.
Dataset Splits No The experiments are numerical comparisons of theoretical bounds for differential privacy, not experiments on a real-world dataset with traditional training/test/validation splits. Therefore, no validation split is explicitly mentioned.
Hardware Specification Yes All the methods are implemented in Python5 and all the experiments are carried out on a Mac Book with 2.5GHz processor and 16GB memory.
Software Dependencies No The paper states 'All the methods are implemented in Python' but does not specify the version number of Python or any other software dependencies with their versions.
Experiment Setup Yes We let the number of compositions n vary from 1 to 10. Since the privacy guarantee decays as n increases and the resulting curves would be very close to the axes, we set θ = 3/ n for the sake of better visibility.