AI Paper Atlas

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

Certified Training: Small Boxes are All You Need

Mark Niklas Mueller, Franziska Eckert, Marc Fischer, Martin Vechev

ICLR 2023 · Adversarial robustness and security › Certified and empirical robustness · approach: Certified training and ensembles · 10 citations (OpenAlex)

Read the paper

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 small, adversarially selected propagation regions with IBP boundsSABR · CIFAR-10, ε∞=2/255standard accuracy: 66.84 -> 79.24Improved
Use small, adversarially selected propagation regions with IBP boundsSABR · CIFAR-10, ε∞=2/255certified accuracy: 52.85 -> 62.84Improved
Use small, adversarially selected propagation regions with IBP boundsSABR · TinyImageNet, ε∞=1/255standard accuracy: 25.92 -> 28.85Improved
Use small, adversarially selected propagation regions with IBP boundsSABR · TinyImageNet, ε∞=1/255certified accuracy: 17.87 -> 20.46Improved
Decrease the subselection ratio λ (ablation)SABR · CIFAR-10standard accuracyImproved
Decrease the subselection ratio λ (ablation)SABR · CIFAR-10, verified with MN-BaBcertified accuracyMixed
Center the propagation region on the original input instead of a high-loss adversarial point (ablation)SABR · CIFAR-10, ε=2/255, first 1000 samplescertified accuracy: 61.0 -> 27.4Hurt
Choose the propagation-region center uniformly at random instead of using an adversarial attack (ablation)SABR · CIFAR-10, ε=2/255, first 1000 samplescertified accuracy: 61.0 -> 28.3Hurt
Use FGSM rather than a strong PGD attack to select the region center (ablation)SABR · CIFAR-10, ε=2/255, first 1000 samplescertified accuracy: 61.0 -> 58.2Hurt
Omit projection that keeps the propagation region inside the adversarial region (ablation)SABR · CIFAR-10, ε=2/255, first 1000 samplesstandard accuracy: 80.4 -> 80.0Hurt
Replace IBP propagation with CROWN-IBP bounds (ablation)SABR · CIFAR-10, ε=2/255, first 1000 samplescertified accuracy: 61.0 -> 56.7Hurt
Use small propagation regions rather than full-region IBP training (ablation)SABR · CIFAR-10, ε=2/255gradient cosine similarity with standard-example loss: 0.5586 -> 0.8071Improved

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