A DIRT-T Approach to Unsupervised Domain Adaptation

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 the violation the cluster assumption; 2) the Decision-boundary Iterative Refinement Training with a Teacher (DIRT-T) model, which takes the VADA model as initialization and employs natural gradient steps to further minimize the cluster assumption violation. Extensive empirical results demonstrate that the combination of these two models significantly improve the state-of-the-art performance on the digit, traffic sign, and Wi-Fi recognition domain adaptation benchmarks.

ICLR 2018

References in corpus (14)

• Instance Normalization: The Missing Ingredient for Fast Stylization
• Learning Transferable Features with Deep Adaptation Networks
• Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results
• Temporal Ensembling for Semi-Supervised Learning
• Domain Separation Networks
• Training Deep Neural Networks on Noisy Labels with Bootstrapping
• Domain Adaptation: Learning Bounds and Algorithms
• Asymmetric Tri-training for Unsupervised Domain Adaptation
• Optimizing Neural Networks with Kronecker-factored Approximate Curvature
• Good Semi-supervised Learning that Requires a Bad GAN
• Virtual Adversarial Training: A Regularization Method for Supervised and Semi-Supervised Learning
• Many Paths to Equilibrium: GANs Do Not Need to Decrease a Divergence At Every Step
• Unsupervised Pixel-Level Domain Adaptation with Generative Adversarial Networks
• Self-ensembling for visual domain adaptation

Cited by in corpus (94)

• A review of domain adaptation without target labels
• Model Adaptation: Unsupervised Domain Adaptation without Source Data
• How Generative Adversarial Networks and Their Variants Work: An Overview
• Maximum Density Divergence for Domain Adaptation
• Tent: Fully Test-time Adaptation by Entropy Minimization
• Uncertainty-aware multi-view co-training for semi-supervised medical image segmentation and domain adaptation
• An Overview of Deep Semi-Supervised Learning
• Unsupervised Domain Adaptation through Self-Supervision
• Semi-supervised semantic segmentation needs strong, varied perturbations
• Do We Really Need to Access the Source Data? Source Hypothesis Transfer for Unsupervised Domain Adaptation
• Constructing Unrestricted Adversarial Examples with Generative Models
• Exploiting the Intrinsic Neighborhood Structure for Source-free Domain Adaptation
• Generalizing to unseen domains via distribution matching
• Self-supervised Autoregressive Domain Adaptation for Time Series Data
• Theoretical Analysis of Self-Training with Deep Networks on Unlabeled Data
• Semi-supervised Domain Adaptation via Minimax Entropy
• Pseudo-Labeling Curriculum for Unsupervised Domain Adaptation
• Unsupervised Domain Adversarial Self-Calibration for Electromyographic-based Gesture Recognition
• Cycle Self-Training for Domain Adaptation
• TimeMatch: Unsupervised Cross-Region Adaptation by Temporal Shift Estimation
• A Survey of Deep Learning for Scientific Discovery
• Frustratingly Simple Domain Generalization via Image Stylization
• Virtual Mixup Training for Unsupervised Domain Adaptation
• Larger Norm More Transferable: An Adaptive Feature Norm Approach for Unsupervised Domain Adaptation
• Augmented Cyclic Adversarial Learning for Low Resource Domain Adaptation
• Semi-supervised Deep Kernel Learning: Regression with Unlabeled Data by Minimizing Predictive Variance
• Constructing Self-motivated Pyramid Curriculums for Cross-Domain Semantic Segmentation: A Non-Adversarial Approach
• Domain-Symmetric Networks for Adversarial Domain Adaptation
• Partial Domain Adaptation without Domain Alignment
• Rethinking Distributional Matching Based Domain Adaptation
• Confidence Regularized Self-Training
• Self-Training and Adversarial Background Regularization for Unsupervised Domain Adaptive One-Stage Object Detection
• Minimum Class Confusion for Versatile Domain Adaptation
• Operational Calibration: Debugging Confidence Errors for DNNs in the Field
• Source Data-absent Unsupervised Domain Adaptation through Hypothesis Transfer and Labeling Transfer
• Unsupervised Domain Adaptation via Structurally Regularized Deep Clustering
• A Prototype-Oriented Framework for Unsupervised Domain Adaptation
• Progressive Feature Alignment for Unsupervised Domain Adaptation
• Harmonizing Transferability and Discriminability for Adapting Object Detectors
• Cluster Alignment with a Teacher for Unsupervised Domain Adaptation
• Vicinal and categorical domain adaptation
• PAC-Bayes and Domain Adaptation
• Semi-supervised Models are Strong Unsupervised Domain Adaptation Learners
• Noise Adaptive Speech Enhancement using Domain Adversarial Training
• OmniNet: A unified architecture for multi-modal multi-task learning
• Unsupervised Traffic Scene Generation with Synthetic 3D Scene Graphs
• Estimating Generalization under Distribution Shifts via Domain-Invariant Representations
• Network Architecture Search for Domain Adaptation
• Label Propagation with Augmented Anchors: A Simple Semi-Supervised Learning baseline for Unsupervised Domain Adaptation
• Angular Visual Hardness
• Unsupervised Domain Adaptation via Structured Prediction Based Selective Pseudo-Labeling
• Improving Unsupervised Domain Adaptation with Variational Information Bottleneck
• Towards Stable and Comprehensive Domain Alignment: Max-Margin Domain-Adversarial Training
• Learning from a Complementary-label Source Domain: Theory and Algorithms
• Learning to Learn Single Domain Generalization
• Domain Adaptation with Auxiliary Target Domain-Oriented Classifier
• A Theory of Label Propagation for Subpopulation Shift
• Contrastively Smoothed Class Alignment for Unsupervised Domain Adaptation
• HoMM: Higher-order Moment Matching for Unsupervised Domain Adaptation
• Butterfly: One-step Approach towards Wildly Unsupervised Domain Adaptation
• DINE: Domain Adaptation from Single and Multiple Black-box Predictors
• Adversarial Appearance Learning in Augmented Cityscapes for Pedestrian Recognition in Autonomous Driving
• Learning Smooth Representation for Unsupervised Domain Adaptation
• A Balanced and Uncertainty-aware Approach for Partial Domain Adaptation
• Self-training Avoids Using Spurious Features Under Domain Shift
• On Deep Domain Adaptation: Some Theoretical Understandings
• Synthetic-to-Real Unsupervised Domain Adaptation for Scene Text Detection in the Wild
• Contradictory Structure Learning for Semi-supervised Domain Adaptation
• Meta Fine-Tuning Neural Language Models for Multi-Domain Text Mining
• SALT: Subspace Alignment as an Auxiliary Learning Task for Domain Adaptation
• Source-Free Adaptation to Measurement Shift via Bottom-Up Feature Restoration
• Towards Fewer Annotations: Active Learning via Region Impurity and Prediction Uncertainty for Domain Adaptive Semantic Segmentation
• Robustified Domain Adaptation
• CUDA: Contradistinguisher for Unsupervised Domain Adaptation
• Incremental Unsupervised Domain-Adversarial Training of Neural Networks
• Integrating Categorical Semantics into Unsupervised Domain Translation
• Domain-Invariant Feature Distillation for Cross-Domain Sentiment Classification
• TOHAN: A One-step Approach towards Few-shot Hypothesis Adaptation
• Composed Fine-Tuning: Freezing Pre-Trained Denoising Autoencoders for Improved Generalization
• Attract or Distract: Exploit the Margin of Open Set
• Metric-Learning-Assisted Domain Adaptation
• On Direct Distribution Matching for Adapting Segmentation Networks
• Matching Embeddings for Domain Adaptation
• Learning Classifiers for Domain Adaptation, Zero and Few-Shot Recognition Based on Learning Latent Semantic Parts
• On Label Shift in Domain Adaptation via Wasserstein Distance
• More Separable and Easier to Segment: A Cluster Alignment Method for Cross-Domain Semantic Segmentation
• PALMAR: Towards Adaptive Multi-inhabitant Activity Recognition in Point-Cloud Technology
• Deep Domain Adaptation under Deep Label Scarcity
• Geometry-Aware Self-Training for Unsupervised Domain Adaptationon Object Point Clouds
• Attention-based Adversarial Appearance Learning of Augmented Pedestrians
• iFAN: Image-Instance Full Alignment Networks for Adaptive Object Detection
• SketchTransfer: A Challenging New Task for Exploring Detail-Invariance and the Abstractions Learned by Deep Networks
• The Role of Embedding Complexity in Domain-invariant Representations
• Unsupervised Domain Adaptation: A Reality Check