qq_32925101的博客

Adversarial Logit PairingAdversarial Logit Pairingwe introduce enhanced defenses using a technique we call logit pairing, a method that encourages logits for pairs of examples to be similar.本文提出了一种logit pairing方法做防御。0. Recall0.1 Adversarial TrainingA

Towards Deep Learning Models Resistant to Adversarial AttacksTowards Deep Learning Models Resistant to Adversarial Attacks (PGD)，ICLR2018，涉及PGD和对抗训练。**Abstract：**本文从优化的角度研究了神经网络的对抗鲁棒性问题。本文提出的方法提供了一个广阔、统一的观点来看待对抗样本的问题。本文提出方法的自然性质使得我们可以可靠的选择训练和攻击神经网络的方法，并且

Adversarial Training for Freeadversarial training for free, NeurIPS 2019Introduction首先是鲁棒性的定义：A robust classifier is one that correctly labels adversarially perturbed images.其次是实现鲁棒性的方法：检测并剔除对抗样本（detecting and rejecting Adv. Examp.）然后是对抗训练的问题，首