BriefGPT.xyz
Jun, 2023
高维回归中基于生成数据的对抗训练：渐进研究
Adversarial Training with Generated Data in High-Dimensional Regression: An Asymptotic Study
HTML
PDF
Yue Xing
TL;DR
本文对高维线性回归中采用伪标签和真实或生成的数据进行的双阶段对抗性训练方法进行了理论分析，证明了该方法可通过适当的L2正则化来避免Ridgeless训练中的双下降现象，从而提高模型性能, 并推导了适用于该方法的快捷交叉验证公式。
Abstract
In recent years, studies such as \cite{carmon2019unlabeled,gowal2021improving,xing2022artificial} have demonstrated that incorporating additional real or generated data with
pseudo-labels
can enhance
adversarial trainin
→