Unsupervised Domain Adaptation through Self-Supervision

2019/09/26 by Yu Sun, Eric Tzeng, Sun, Yu +5 · 169 citations
Computer Science · Mathematics · Neuroscience · Psychology · #Adaptation (eye) #Artificial intelligence #Computer Vision and Pattern Recognition (cs.CV) #Computer science #Domain (mathematical analysis) #Domain Adaptation and Few-Shot Learning #Domain adaptation #FOS: Computer and information sciences #Human Pose and Action Recognition #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Mathematics #Multimodal Machine Learning Applications #Neuroscience #Psychology #cs.CV #cs.LG #stat.ML

paper · pdf · doi:10.48550/arxiv.1909.11825

published in arXiv (Cornell University) (Cornell University)

openalex publication_date 2019/09/26 · arxiv created 2019/09/29 · arxiv updated 2019/10/01 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

This paper addresses unsupervised domain adaptation, the setting where labeled training data is available on a source domain, but the goal is to have good performance on a target domain with only unlabeled data. Like much of previous work, we seek to align the learned representations of the source and target domains while preserving discriminability. The way we accomplish alignment is by learning to perform auxiliary self-supervised task(s) on both domains simultaneously. Each self-supervised task brings the two domains closer together along the direction relevant to that task. Training this jointly with the main task classifier on the source domain is shown to successfully generalize to the unlabeled target domain. The presented objective is straightforward to implement and easy to optimize. We achieve state-of-the-art results on four out of seven standard benchmarks, and competitive results on segmentation adaptation. We also demonstrate that our method composes well with another popular pixel-level adaptation method.

Citations

• Gradient-based learning applied to document recognition
• Domain Separation Networks
• Unsupervised Cross-Domain Image Generation
• NIPS 2016 Tutorial: Generative Adversarial Networks
• Domain Adaptation for Visual Applications: A Comprehensive Survey
• Asymmetric Tri-training for Unsupervised Domain Adaptation
• Geometry of Optimization and Implicit Regularization in Deep Learning
• Stabilizing GAN Training with Multiple Random Projections
• Rethinking Atrous Convolution for Semantic Image Segmentation
• Many Paths to Equilibrium: GANs Do Not Need to Decrease a Divergence At Every Step
• CyCADA: Cycle-Consistent Adversarial Domain Adaptation
• Stronger generalization bounds for deep nets via a compression approach
• A DIRT-T Approach to Unsupervised Domain Adaptation
• What is Local Optimality in Nonconvex-Nonconcave Minimax Optimization?
• Colorization as a Proxy Task for Visual Understanding
• Understanding deep learning requires rethinking generalization
• Boosting Supervision with Self-Supervision for Few-shot Learning
• Learning Multiple Layers of Features from Tiny Images

Cited by

• Test-Time Warmup for Multimodal Large Language Models
• SynthGenNet: a self-supervised approach for test-time generalization using synthetic multi-source domain mixing of street view images
• Neutralizing Token Aggregation via Information Augmentation for Efficient Test-Time Adaptation
• Box Pose and Shape Estimation and Domain Adaptation for Large-Scale Warehouse Automation
• FreeDNA: Endowing Domain Adaptation of Diffusion-Based Dense Prediction with Training-Free Domain Noise Alignment
• Rethinking Distributional Matching Based Domain Adaptation
• Auxiliary Signal-Guided Knowledge Encoder-Decoder for Medical Report Generation
• BoTTA: Benchmarking on-device Test Time Adaptation

Related

• Unsupervised Domain Adaptation in Semantic Segmentation Based on Pixel Alignment and Self-Training
• Cross-domain Self-supervised Learning for Domain Adaptation with Few Source Labels
• Self-Supervised Domain Adaptation with Consistency Training
• Tackling Partial Domain Adaptation with Self-Supervision
• Unsupervised Domain Adaptation: A Multi-task Learning-based Method