AI Paper Atlas

Research map / Adversarial robustness and security / Certified and empirical robustness

BREAKING CERTIFIED DEFENSES: SEMANTIC ADVERSARIAL EXAMPLES WITH SPOOFED ROBUSTNESS CERTIFICATES

Amin Ghiasi, Ali Shafahi, Tom Goldstein

ICLR 2020 · Adversarial robustness and security › Certified and empirical robustness · approach: Provable defenses and their limits · 16 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
Use Shadow Attack rather than natural inputsRandomized Smoothing on CIFAR-10 · CIFAR-10, σ=0.12–1.00certified radius: 0.14 -> 0.22; 0.30 -> 0.35; 0.47 -> 0.65; 0.78 -> 0.85Improved
Use Shadow Attack rather than natural inputsRandomized Smoothing on ImageNet · ImageNet, σ=0.25–1.00certified radius: 0.30 -> 0.31; 0.61 -> 0.38; 1.04 -> 0.64Mixed
Use Shadow Attack on adversarially trained smoothingAdversarially trained Randomized Smoothing · CIFAR-10, σ=0.5certified radius: 0.60 -> 0.65Improved
Use Shadow Attack images instead of natural imagesCROWN-IBP, 9 small models · CIFAR-10, ε=2/255error rate: 57.55 -> 53.89Improved
Use Shadow Attack images instead of natural imagesCROWN-IBP, 8 large models · CIFAR-10, ε=2/255error rate: 53.9 -> 49.77Improved
Use Shadow Attack images instead of natural imagesCROWN-IBP, 9 small models · CIFAR-10, ε=8/255error rate: 72.15 -> 66.94Improved
Use Shadow Attack images instead of natural imagesCROWN-IBP, 8 large models · CIFAR-10, ε=8/255error rate: 71.17 -> 67.32Improved
Constrain the perturbation to one shared grayscale channel rather than using three channels (ablation)Shadow Attack · CIFAR-10 validation examples where smoothing does not abstainattack success rate: 98.5%No clear effect
Increase the number of optimization steps to 10 (ablation)Shadow Attack · CIFAR-10optimization convergence: 10 stepsImproved
Increase optimization steps from 10 to 300 (ablation)Shadow Attack · CIFAR-10 randomized smoothing and CROWN-IBP attacksattack performanceNo clear effect
Vary the channel-dissimilarity penalty λs (ablation)Three-channel Shadow Attack · Tiny-CIFAR-10perturbation channel dissimilarityImproved
Vary the total-variation penalty λtv (ablation)Shadow Attack · Tiny-CIFAR-10perturbation smoothnessImproved

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