CEAR is a hybrid ensemble method that adds per-network Gaussian noise and temperature scaling, uses two voting schemes on noisy logits, and extends randomized smoothing to deliver certified robustness on MNIST, CIFAR-10 and TinyImageNet.
FailureCasesAreBetterLearnedButBoundarySaysSorry:Facilitating Smooth Perception Change for Accuracy-Robustness Trade-Off in Adversarial Training
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
CEAR: Certified Ensemble Adversarial Robustness in DNNs
CEAR is a hybrid ensemble method that adds per-network Gaussian noise and temperature scaling, uses two voting schemes on noisy logits, and extends randomized smoothing to deliver certified robustness on MNIST, CIFAR-10 and TinyImageNet.