Domain Adaptations for Computer Vision Applications
Oscar Beijbom, Department of Computer Science and Engineering, University of California, San Diego
Abstract
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.
中文速览
训练数据和测试数据来自不同分布这一现实挑战，催生了领域自适应（Domain Adaptation）这一研究方向：如何借助标注充足的"源域"数据，帮助模型在标注稀缺的"目标域"上做出准确预测。这篇综述系统梳理了解决该问题的三大类方法——实例加权（通过重新赋予源域样本权重来弥合分布差异）、将源域分布用作目标域模型的先验正则（包括贝叶斯先验和判别式先验两种思路）、以及寻找跨域公共表示（将两个域的特征映射到同一空间以对齐分布），并以计算机视觉领域的最新进展为重点，同时涉及自然语言处理、语音等应用场景。文章还厘清了领域自适应与迁移学习、协变量偏移、主动学习等近邻领域之间的联系与区别。这项综述的价值在于，它为一个定义尚不统一、方法流派纷繁的新兴领域提供了统一的符号体系和清晰的分类框架，为后续研究者快速把握全局提供了有用的导航图。
原文 arXiv:1211.4860;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1211.4860v1