Research map / Label-efficient and robust learning / Domain adaptation and generalization
A DIRT-T Approach to Unsupervised Domain Adaptation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| refine VADA with DIRT-T | VADA · Visual domain adaptation benchmarks | accuracy: DIRT-T consistently improves upon VADA | Helped |
| refine VADA with DIRT-T | VADA · Room A → Room B | accuracy: 53.0 -> 53.0 | No clear effect |
| add virtual adversarial training to VADA (ablation) | VADA · Visual domain adaptation benchmarks | accuracy | Helped |
| add virtual adversarial training during DIRT-T refinement (ablation) | DIRT-T initialized from VADA no-vat · Visual domain adaptation benchmarks | accuracy: 94.8 -> 98.3; 96.3 -> 99.4; 68.6 -> 69.8; 94.4 -> 95.3; 99.1 -> 99.6; 69.2 -> 71.0 | Helped |
| remove virtual adversarial training from VADA (ablation) | VADA · Visual domain adaptation benchmarks | accuracy: 93.8, 83.1, 66.8, 93.4, 98.4, 79.1, 68.6 | Mixed |
| remove the teacher KL constraint during DIRT-T (ablation) | DIRT-T · SVHN → MNIST and STL → CIFAR | accuracy | Hurt |
| apply instance normalization to inputs (ablation) | DIRT-T · MNIST → SVHN | accuracy: 54.5 -> 76.5 | Helped |
| apply instance normalization to inputs (ablation) | DIRT-T · SYN DIGITS → SVHN | accuracy: 96.1 -> 96.2 | No clear effect |
| apply instance normalization to inputs (ablation) | DIRT-T · STL → CIFAR | accuracy: 75.3 -> 73.3 | Hurt |
| replace natural-gradient refinement with naive gradient descent (remove KL constraint) (ablation) | DIRT-T · STL → CIFAR | accuracy | Hurt |
| replace natural-gradient refinement with naive gradient descent (remove KL constraint) (ablation) | DIRT-T · SVHN → MNIST | accuracy | Mixed |
| apply DIRT-T refinement to VADA (ablation) | VADA · MNIST → SVHN | accuracy: 47.5 -> 54.5; 73.3 -> 76.5 | Helped |
Closest papers by meaning
- Adversarial Discriminative Domain Adaptation (CVPR 2017 · similarity 0.91)
- Transferable Adversarial Training: A General Approach to Adapting Deep Classifiers (ICML 2019 · similarity 0.91)
- Domain-Symmetric Networks for Adversarial Domain Adaptation (CVPR 2019 · similarity 0.90)
- f-Domain Adversarial Learning: Theory and Algorithms (ICML 2021 · similarity 0.90)
- Discriminative Adversarial Domain Adaptation (AAAI 2020 · similarity 0.90)
- Can We Evaluate Domain Adaptation Models Without Target-Domain Labels? (ICLR 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.