BriefGPT.xyz
Oct, 2019
实例自适应对抗训练：神经网络中更好的准确性和权衡
Instance adaptive adversarial training: Improved accuracy tradeoffs in neural nets
HTML
PDF
Yogesh Balaji, Tom Goldstein, Judy Hoffman
TL;DR
本文提出了一种实例自适应对抗训练技术，通过在每个训练样本周围施加特定的扰动边缘来实现更好的泛化能力和测试准确性。
Abstract
adversarial training
is by far the most successful strategy for improving robustness of
neural networks
to adversarial attacks. Despite its success as a defense mechanism,
→