AI Paper Atlas

Research map / Federated learning and privacy / Differential privacy

Numerical Composition of Differential Privacy

Sivakanth Gopi, Yin Tat Lee, Lukas Wutschitz

NeurIPS 2021 · Federated learning and privacy › Differential privacy · approach: Privacy accounting and mechanisms · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
truncate PRV domain at L = Theta(1) using tight tail bound instead of O~(sqrt(k))PRV composition algorithm · DP-SGD privacy curverunning time and memory: O~(sqrt(k)) -> O~(1)Improved
proposed algorithm versus prior FFT algorithm of Koskela et al. at aligned accuracyPRV composition algorithm · DP-SGD sigma=0.8, p=4e-3, eps=1.5, varying stepsruntimeImproved
large number of DP-SGD steps (beyond 2,000) with the prior methodprior FFT algorithm of Koskela et al. · DP-SGD, sigma=0.8, p=4e-3numerical stability: unstable beyond 2,000 stepsHurt
number of discretisation points needed for convergence: proposed vs prior methodPRV composition algorithm · p=4e-3, sigma=0.8, eps=1.5, k=1000error of delta vs number of discretisation pointsImproved
compare error bounds at the same mesh size 8e-4 for k = 10, 100, 1000PRV composition algorithm vs prior method · p=1e-2, sigma=0.8width of delta error bounds: prior method reasonable at small k, worse at k=1000Mixed
use PRV composition bounds versus GDP and RDP accountantsDP-SGD privacy accounting · DP-SGD privacy curveprivacy curve boundsImproved
validate against exact analytical Gaussian curve at p=1PRV composition algorithm · p=1 (Gaussian mechanism)sandwich of true delta(eps): delta_error 10^-10Improved
push delta below floating point resolutionPRV composition algorithm · p=1 analytical comparison, ~10^4 discretization pointsbound meaningfulness: fails for delta < 2 x 10^-11Hurt
analysis via coupling approximation instead of convolution stabilityerror analysis · runtime dependence on delta_errorruntime dependence: 1/delta_error -> sqrt(log(1/delta_error))Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy