BriefGPT.xyz
Aug, 2022
基于次最终激活的对抗性训练的领域自适应
Domain Adaptation with Adversarial Training on Penultimate Activations
HTML
PDF
Tao Sun, Cheng Lu, Haibin Ling
TL;DR
以对抗训练为基础，使用最后一层线性分类器的激活输入特征进行无监督领域自适应预测，结合激活归一化技术，提高模型预测能力。
Abstract
Enhancing model prediction confidence on unlabeled target data is an important objective in
unsupervised domain adaptation
(UDA). In this paper, we explore
adversarial training
on penultimate activations, ie, inp
→