cifar10-297fb111·1 events·first seen Aliases: CIFAR10
A new arXiv preprint introduces a certified training method that encodes convolutional perturbations (e.g., motion blur) to train vision models with formal robustness guarantees. The approach outperforms adversarial training, achieving over 80% robust accuracy against motion blur on CIFAR10 while preserving standard accuracy. The work targets safety-critical deployment scenarios where empirical robustness methods like data augmentation are insufficient due to lack of formal guarantees.