Numerical Composition of Differential Privacy

Authors: Sivakanth Gopi, Yin Tat Lee, Lukas Wutschitz

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the utility of our composition method by computing the privacy curves for the DP-SGD algorithm, which is one of the most important algorithms in differential privacy. Figure 1: Case study on DP-SGD. Sampling probability p = 10 3, noise scale σ = 0.8, δ = 10 7.
Researcher Affiliation Collaboration Sivakanth Gopi Microsoft Research sigopi@microsoft.com Yin Tat Lee University of Washington yintat@uw.edu Lukas Wutschitz Microsoft lukas.wutschitz@microsoft.com
Pseudocode Yes Algorithm 1: Compose PRV: Composing privacy curves using PRVs. Algorithm 2: Discretize PRV: Discretize and truncate a PRV
Open Source Code Yes Code is available at https://github.com/microsoft/prv_accountant.
Open Datasets No The paper discusses the DP-SGD algorithm and its parameters used in experiments but does not provide specific access information (link, DOI, repository, or formal citation) for any publicly available dataset used for their empirical evaluation.
Dataset Splits No The paper does not provide specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) needed to reproduce the data partitioning.
Hardware Specification Yes All experiments are performed on a Intel Xeon W-2155 CPU with 3.30GHz with 128GB of memory.
Software Dependencies No The paper mentions the use of Fast Fourier Transform (FFT) and references an implementation from prior work, but it does not provide specific software dependencies with version numbers (e.g., library names like PyTorch or NumPy along with their versions).
Experiment Setup Yes Sampling probability p = 10 3, noise scale σ = 0.8, δ = 10 7.