We consider a model of robust learning in an adversarial environment. The learner gets
uncorrupted training data with access to possible corruptions that may be effected by the …

Recent studies demonstrate that deep networks, even robustified by the state-of-the-art
adversarial training (AT), still suffer from large robust generalization gaps, in addition to the …

Several recent results provide theoretical insights into the phenomena of adversarial
examples. Existing results, however, are often limited due to a gap between the simplicity of …

Network pruning has been known to produce compact models without much accuracy
degradation. However, how the pruning process affects a network's robustness and the …

We first elucidate various fundamental properties of optimal adversarial predictors: the
structure of optimal adversarial convex predictors in terms of optimal adversarial zero-one …

We consider a model of robust learning in an adversarial environment. The learner gets
uncorrupted training data with access to possible corruptions that may be affected by the …

Benign overfitting is the phenomenon wherein none of the predictors in the hypothesis class
can achieve perfect accuracy (ie, non-realizable or noisy setting), but a model that …

Abstract Machine learning techniques were initially designed for stationary and benign
environments, where the training and test data are assumed to be generated from the same …

Adversarial training has emerged as a popular approach for training models that are robust
to inference-time adversarial attacks. However, our theoretical understanding of why and …

In dieser Arbeit untersuchen wir die Robustheit und Verallgemeinerungseigenschaften von
Deep Neural Networks (DNNs) unter verschiedenen rauschbehafteten Bedingungen, die …