BriefGPT.xyz
Feb, 2020
对抗鲁棒深度学习中的过拟合问题
Overfitting in adversarially robust deep learning
HTML
PDF
Leslie Rice, Eric Wong, J. Zico Kolter
TL;DR
研究深度学习领域中常用的过参数化网络和尽可能训练的现象,发现对于对抗训练的深度网络来说过拟合确实会对其稳健性产生很大的负面影响,因此建议使用提前停止等方法来取得相似的性能提升。
Abstract
It is common practice in
deep learning
to use
overparameterized networks
and train for as long as possible; there are numerous studies that show, both theoretically and empirically, that such practices surprising
→