AI Paper Atlas

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

Understanding Deep Gradient Leakage via Inversion Influence Functions

Haobo Zhang, Junyuan Hong, Yuyang Deng, Mehrdad Mahdavi, Jiayu Zhou

NeurIPS 2023 · Federated learning and privacy › Privacy attacks and secure learning · approach: Gradient inversion attacks · 2 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
Estimate recovery risk with I²F power iteration instead of directly minimizing inversion lossCIFAR10, ResNet18 · CIFAR10computation time: 5 seconds at most; 150 secondsImproved
Perturb gradients along Jacobian directions with smaller eigenvaluesDGL defenses · MNIST and CIFAR10recovery errorImproved
Use homogeneous Gaussian gradient noiseDGL defenses · MNIST and CIFAR10recovery error across samples and classesMixed
Initialize with Kaiming or Xavier instead of uniformLeNet · MNIST and CIFAR10recovery error under Gaussian gradient noiseImproved
Initialize with normal rather than Kaiming or XavierLeNet · MNIST and CIFAR10recovery error under Gaussian gradient noiseHurt
Use the I²F lower bound rather than the numerator-only approximation (ablation)I²F · Different models and datasetsapproximation of recovery errorImproved
Use I²F to estimate recovery error across model and dataset settingsDGL and GS attacks · MNIST and CIFAR10correlation with recovery errorMixed
Apply I²F to text-gradient inversionBERT and GPT-2 with TAG · WikiText-103correlation with text similarity metricsImproved

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