AI Paper Atlas

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

Boosting Randomized Smoothing with Variance Reduced Classifiers

Miklós Z. Horváth, Mark Niklas Mueller, Marc Fischer, Martin Vechev

ICLR 2022 · Adversarial robustness and security › Certified and empirical robustness · approach: Randomized smoothing certification · 8 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
Average logits from an ensemble of classifiers rather than use one classifierRandomized Smoothing classifiers · CIFAR10 and ImageNetACR: 5\% to 21\%Improved
Increase ensemble size (ablation)Randomized Smoothing classifiers · CIFAR10ACRImproved
Use ensembles of 3–5 models rather than a single model (ablation)Randomized Smoothing classifiers · CIFAR10ACRImproved
Use 3 ResNet20 models instead of one ResNet110 (ablation)Gaussian-trained Randomized Smoothing · CIFAR10ACRImproved
Use 5 ResNet20 models instead of one ResNet110 (ablation)Consistency-trained Randomized Smoothing · CIFAR10ACRImproved
Increase perturbation samples for a single classifier instead of using an ensemble (ablation)Randomized Smoothing classifier · CIFAR10ACRMixed
Combine classifiers trained with different methods (ablation)Randomized Smoothing ensemble · CIFAR10ACR: up to 5\%Improved
Use denoised classifiers in the ensemble (ablation)Denoised Randomized Smoothing · CIFAR10ACR: up to 17\%Improved
Apply adaptive sampling for predetermined certification radiiRandomized Smoothing certification · CIFAR10SampleRF: up to 55-foldImproved
Use exponentially increasing adaptive-sampling phases, particularly 100, 1'000, 10'000, 120'000 (ablation)CertifyAdp · CIFAR10SampleRF: 33.91Improved
Use K-Consensus with K=2 instead of full evaluation of a 10-model ensemble (ablation)Consistency ResNet110 ensemble · CIFAR10, σ=0.25ACR: 0.546 -> 0.576Improved
Use K-Consensus with K=5 instead of full evaluation of a 10-model ensemble (ablation)Consistency ResNet110 ensemble · CIFAR10, σ=0.25ACR: 0.583 -> 0.583No clear effect

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