Do all adversarial examples have the same consequences? An autonomous driving
system misclassifying a pedestrian as a car may induce a far more dangerous --
and even potentially lethal -- behavior than, for instance, a car as a bus. In
order to better tackle this important problematic, we introduce the concept of
hierarchical adversarial robustness. Given a dataset whose classes can be
grouped into coarse-level labels, we define hierarchical adversarial examples
as the ones leading to a misclassification at the coarse level. To improve the
resistance of neural networks to hierarchical attacks, we introduce a
hierarchical adversarially robust (HAR) network design that decomposes a single
classification task into one coarse and multiple fine classification tasks,
before being specifically trained by adversarial defense techniques. As an
alternative to an end-to-end learning approach, we show that HAR significantly
improves the robustness of the network against $\ell_2$ and $\ell_{\infty}$
bounded hierarchical attacks on the CIFAR-10 and CIFAR-100 dataset.

介绍了一种分类网络设计 -- 分层敌对强韧网络（HAR），将分类任务分解成一个粗分类和多个细分类任务，在经过敌对防御技术的有针对性训练后，极大地提高了神经网络对层次攻击的强度，实验表明对 CIFAR 数据集 L2 和 L∞有显著改进。

深层神经网络的层级对抗鲁棒性的提升

Improving Hierarchical Adversarial Robustness of Deep Neural Networks

The network-based machine learning algorithm is very powerful tools. However,
it requires huge training dataset. Researchers often meet privacy issues when
they collect image dataset especially for surveillance applications. A
learnable image encryption scheme is introduced. The key idea of this scheme is
to encrypt images, so that human cannot understand images but the network can
be train with encrypted images. This scheme allows us to train the network
without the privacy issues. In this paper, a simple learnable image encryption
algorithm is proposed. Then, the proposed algorithm is validated with cifar
dataset.

介绍了一种可学习的图像加密方案，用于保护图像隐私，在不侵犯隐私的前提下，通过加密的图像进行网络训练并得到验证。