CC-CERT: A Probabilistic Approach to Certify General Robustness of Neural Networks

In safety-critical machine learning applications, it is crucial to defend models against adversarial attacks --- small modifications of the input that change predictions. Besides rigorously studied $\ell_p$-bounded additive perturbations, semantic perturbations (e.g. rotation, translation) raise a serious concern on deploying ML systems in real-world. Therefore, important provide provable guarantees for deep semantically meaningful transformations. this paper, we propose new universal probabilistic certification approach based Chernoff-Cramer bounds can be used general attack settings. We estimate probability model fail if sampled from certain distribution. Our theoretical findings are supported by experimental results different datasets.