BriefGPT.xyz
Mar, 2023
认证鲁棒神经网络: 泛化和抗污染性
Certified Robust Neural Networks: Generalization and Corruption Resistance
HTML
PDF
Amine Bennouna, Ryan Lucas, Bart Van Parys
TL;DR
通过提出综合强健的（HR）训练程序，理论和实验证明了其在对抗错误损失方面具有SOTA性能，并克服了在敌对训练中普遍存在的过度拟合问题。
Abstract
adversarial training
aims to reduce the problematic susceptibility of modern
neural networks
to small data perturbations. Surprisingly,
overfitti
→