AI Paper Atlas

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

Dual Randomized Smoothing: Beyond Global Noise Variance

Chenhao Sun, Yuhao Mao, Martin Vechev

ICLR 2026 · Adversarial robustness and security › Certified and empirical robustness · approach: Randomized smoothing certification · 0 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
replace global sigma with input-dependent sigma from variance estimator (off-the-shelf classifier)diffusion denoised smoothing · CIFAR-10certified accuracy (radius 1.0): 0.0 (sigma=0.25) / 19.34 (sigma=1.0) -> 19.85 at 1.0; but 86.61 -> 68.34 at radius 0Improved
use Dual RS instead of Multiscale (classifier fine-tuned)Dual RS · CIFAR-10certified accuracy (radius 0.5): 39.15 -> 45.27Improved
use Dual RS instead of MultiscaleDual RS · CIFAR-10certified accuracy (radius 0.0): 76.51 -> 70.53Hurt
use Dual RS instead of MultiscaleDual RS · ImageNetcertified accuracy (radius 0.5): 55.8 -> 60.6Improved
use Dual RS instead of MultiscaleDual RS · ImageNetcertified accuracy (radius 2.0): 25.2 -> 17.0Hurt
fine-tune the classifier under estimated variances (one round) (ablation)Dual RS · CIFAR-10certified accuracy (radius 0.5): 41.28 -> 45.27Improved
use soft cross-entropy instead of standard CE for variance estimator (ablation)Dual RS · CIFAR-10accuracy-radius curveImproved
add consistency loss to variance estimator (ablation)Dual RS · CIFAR-10accuracy-radius curveMixed
train variance estimator with much smaller N or train subset (ablation)Dual RS · CIFAR-10certified accuracyNo clear effect
change variance estimator architecture (ablation)Dual RS · CIFAR-10certified accuracyNo 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