AI Paper Atlas

Research map / Federated learning and privacy / Privacy attacks and secure learning

Bounding training data reconstruction in DP-SGD

Jamie Hayes, Borja Balle, Saeed Mahloujifar

NeurIPS 2023 · Federated learning and privacy › Privacy attacks and secure learning · approach: Membership inference and auditing

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
use the new ReRo bound instead of prior RDP-based boundDP-SGD reconstruction upper bound · advantage bound, prior size 100, sigma=1advantage upper bound: 0.346 -> 0.077Improved
use the new ReRo bound instead of prior Monte-Carlo boundDP-SGD reconstruction upper bound · prior size 10, sigma=0.5advantage upper bound: 0.771 -> 0.737Improved
increase noise sigmanew ReRo upper bound · prior size 10advantage upper bound: 0.737 -> 0.080Improved
replace gradient-based prior-oblivious attack with prior-aware attackreconstruction attack · CIFAR-10reconstruction successImproved
increase size of the priorupper bound and attack · MNISTreconstruction successHurt
reduce sampling rate q from 0.99 to 0.01 at fixed eps=4upper bound · MNISTreconstruction upper bound: 0.35 -> 0.20Improved
reduce sampling rate q from 0.99 to 0.01 at fixed eps=4improved prior-aware attack · MNISTattack success: 0.32 -> 0.15Improved
use improved attack using only top qT inner products (ablation)prior-aware attack · MNIST q=0.02attack successImproved
increase T or qupper bound · DP-SGDreconstruction upper boundHurt
use eps in {10,100} with RDP-based boundprior RDP bound · CIFAR-10boundHurt
larger model size (7,960 vs 36.5M parameters)reconstruction attack tightness · MNIST vs CIFAR-10attack tightnessNo clear effect
more samples for gamma estimationAlgorithm 1 · gamma stdstd of gamma: <0.019 -> <0.0017Improved

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