A DIRT-T Approach to Unsupervised Domain Adaptation
Rui Shu†, Hung H. Bui‡, Hirokazu Narui†,、Stefano Ermon† †Stanford University ‡DeepMind Work was done during first author’s internship at Adobe Research.
Abstract
Domain adaptation refers to the problem of leveraging labeled data in a source domain to learn an accurate model in a target domain where labels are scarce or unavailable. A recent approach for finding a common representation of the two domains is via domain adversarial training (Ganin & Lempitsky, 2015), which attempts to induce a feature extractor that matches the source and target feature distributions in some feature space. However, domain adversarial training faces two critical limitations: 1) if the feature extraction function has high-capacity, then feature distribution matching is a weak constraint, 2) in non-conservative domain adaptation (where no single classifier can perform well in both the source and target domains), training the model to do well on the source domain hurts performance on the target domain. In this paper, we address these issues through the lens of the cluster assumption, i.e., decision boundaries should not cross high-density data regions. We propose two novel and related models: 1) the Virtual Adversarial Domain Adaptation (VADA) model, which combines domain adversarial training with a penalty term that punishes violation of the cluster assumption; 2
中文速览
领域自适应（domain adaptation）要解决的是：源域有大量标注数据，而目标域标注稀缺甚至完全没有，如何让模型在目标域也表现良好。现有的域对抗训练方法试图让特征提取器对两个域输出相似的分布，但存在两个硬伤：特征提取器容量足够大时，分布匹配约束形同虚设；而在"非保守"场景下（两域最优分类器本就不同），强行在源域表现好反而会损害目标域性能。为此，本文引入"聚类假设"——决策边界不应穿越高密度数据区域——提出了两个新模型：VADA 在域对抗训练基础上加入虚拟对抗训练（virtual adversarial training）和条件熵最小化，迫使决策边界远离目标域数据；DIRT-T 则以 VADA 为初始化，用自然梯度步进一步精细调整，专注于减少目标域上的聚类假设违反。在数字识别、交通标志识别和 Wi-Fi 活动识别等多个基准上，两者组合均刷新了最优性能，在极具挑战性的 MNIST→SVHN 任务上更比此前最佳方法高出 20% 以上，为深度神经网络时代的无监督域自适应提供了一套更可靠的解决方案。
原文 arXiv:1802.08735;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1802.08735v2