Application of Metamaterial in Vibration Suppression of Rotor System

Abstract

Deep neural networks have been widely used in various downstream tasks, especially those safety-critical scenario such as autonomous driving, but deep networks are often threatened by adversarial samples [1]. Such adversarial attacks can be invisible to human eyes, but can lead to DNN misclassification, and often exhibits transferability between deep learning and machine learning models [2] and real-world achievability [3].Adversarial attacks can be divided into white-box attacks (Section 2.1), for which the attacker knows the parameters and gradient of the model, and black-box attacks (Section 2.2), for the latter, the attacker can only obtain the input and output of the model. In terms of the attacker's purpose, it can be divided into targeted attacks and non-targeted attacks, which means that the attacker wants the model to misclassify the original sample into the specified class, which is more practical, while the non-targeted attack just needs to make the model misclassify the sample. The black box setting is a scenario we will encounter in practice.Black-box attacks can also be divided into query-based attacks, which require a lot of repeated query model output to adjust perturbations, while transfer-based attacks do not, which makes the latter easier to do because too many queries are not allowed in practice.Transferbased attacks often require the use of a white-box surrogate model to create adversarial perturbations, which are mostly developed from existing white-box attacks.