m0_51496961的博客

Progressive Feature Alignment for Unsupervised Domain Adaptation用于无监督领域自适应的渐进式特征对齐Published in: 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)概览UDA： 标签丰富的源域到一个完全未标记的目标域。Recent Approaches： 伪标签进行判别性域迁移，以强制跨源域和目标域的类级(class-lev

Asymmetric Tri-training for Unsupervised Domain Adaptation  深层模型需要大量的标注样本进行训练，但是收集不同领域的标注样本是代价昂贵的。  无监督领域自适应：利用标注的有标注的源样本和目标样本训练一个在目标域上能够很好地work的分类器。存在的问题虽然很多的方法去对齐源域和目标域样本的分布，但是简单地匹配分布不能确保目标域上的准确率。为了学习目标域的判别性的表示，假设人工标记的目标样本可以得到很好的表示。==Tri-training：==

Normalized Wasserstein for Mixture Distributions with Applications in Adversarial Learning and Domain AdaptationAbstractIntroductionNormalized Wasserstein MeasureNormalized Wasserstein in Domain AdaptationUDA for supervised tasks*MNIST → MNIST-M*VISDA*Mode

Unsupervised Domain Adaptation via Structured Prediction Based Selective Pseudo-Labeling基于结构化预测的选择性伪标记的无监督域自适应AbstractIntroductionRelated WorkProposed MethodExperiments and Results基于结构化预测的选择性伪标记的无监督域自适应AbstractIntroductionRelated WorkProposed MethodE

【论文笔记】用于无监督领域自适应的图卷积对抗网络GCAN: Graph Convolutional Adversarial Network for Unsupervised Domain Adaptation