Domain Adaptations for Computer Vision Applications

中文速览

当训练数据和实际测试数据来自不同分布、目标领域又缺少标注时，如何借助源领域的丰富标注数据做好目标领域分类，是领域自适应要解决的问题。文章系统梳理了这一方向，按思路归纳为给源样本加权、用源领域模型约束目标模型，以及学习能对齐源域和目标域的共同特征表示，并区分了监督、半监督和无监督等设置。已有方法分别利用类别比例或特征分布差异修正训练权重，或通过贝叶斯先验、支持向量机等方式迁移源领域知识，覆盖计算机视觉、自然语言处理、语音等应用。这个综述的重要性在于厘清了领域自适应与迁移学习、协变量偏移等相关概念，为标注昂贵、数据分布不断变化的实际任务提供了方法地图。

摘要

A basic assumption of statistical learning theory is that train and test data are drawn from the same underlying distribution. Unfortunately, this assumption doesn’t hold in many applications. Instead, ample labeled data might exist in a particular ‘source’ domain while inference is needed in another, ‘target’ domain. Domain adaptation methods leverage labeled data from both domains to improve classification on unseen data in the target domain. In this work we survey domain transfer learning methods for various application domains with focus on recent work in Computer Vision.

术语表

statistical learning theory

统计学习理论

domain adaptation (DA)

领域自适应

domain transfer learning

领域迁移学习

transfer learning (TL)

迁移学习

source domain

源领域

target domain

目标领域

source distribution

源分布

target distribution

目标分布

joint probability distribution

联合概率分布

marginal probability distribution

边缘概率分布

covariate shift

协变量偏移

active learning

主动学习

human-in-the-loop

人在回路

crowdsourcing

众包

weakly supervised learning

弱监督学习

multiple instance learning

多示例学习

latent structural SVMs

潜在结构 SVM

semi-supervised learning

半监督学习

co-training

协同训练

one-shot learning

单样本学习

model adaptation

模型自适应

Adapted Gaussian Mixture Models

自适应高斯混合模型

cross-modal classification

跨模态分类

cross-modal retrieval

跨模态检索

instance weighting

实例加权

empirical risk minimization

经验风险最小化

loss function

损失函数

supervised domain adaptation

有监督领域自适应

unsupervised domain adaptation

无监督领域自适应

semi-supervised domain adaptation

半监督领域自适应