BriefGPT.xyz
May, 2020
使用Lipschitz界限训练鲁棒神经网络
Training robust neural networks using Lipschitz bounds
HTML
PDF
Patricia Pauli, Anne Koch, Julian Berberich, Frank Allgöwer
TL;DR
通过设计一种基于交替方向乘子法的最优化方案来训练多层神经网络,同时鼓励通过保持其利普希茨常数来促进鲁棒性,从而解决基于输入的扰动的效应以及提高神经网络的鲁棒性。该文设计了两个训练程序,最终提供了两个例子来证明这种方法成功地提高了神经网络的鲁棒性。
Abstract
Due to their susceptibility to
adversarial perturbations
,
neural networks
(NNs) are hardly used in safety-critical applications. One measure of
r
→