Research map / Label-efficient and robust learning / Domain adaptation and generalization
DeepJDOT: Deep Joint Distribution Optimal Transport for Unsupervised Domain Adaptation
Bharath Bhushan Damodaran, Benjamin Kellenberger, Remi Flamary, Devis Tuia, Nicolas Courty
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| compute optimal transport coupling in deep CNN feature layer instead of image space (ablation) | JDOT/DeepJDOT · MNIST -> MNIST-M | target accuracy: 66.7 -> 92.4 | Improved |
| compute coupling in deep features instead of image space (ablation) | JDOT/DeepJDOT · SVHN -> MNIST | target accuracy: 67.6 -> 96.7 | Improved |
| compute coupling in deep features instead of image space (ablation) | JDOT/DeepJDOT · USPS -> MNIST | target accuracy: 90.5 -> 96.4 | Improved |
| add source classification loss Ls to the JDOT loss (ablation) | DeepJDOT · USPS -> MNIST | target accuracy: 86.41 -> 96.4 | Improved |
| add source classification loss Ls to the JDOT loss (ablation) | DeepJDOT · MNIST -> MNIST-M | target accuracy: 73.6 -> 92.4 | Improved |
| add target pseudo-label transport loss Lt (ablation) | DeepJDOT (Ls + alpha d) · MNIST -> MNIST-M | target accuracy: 82.3 -> 92.4 | Improved |
| add target pseudo-label transport loss Lt (ablation) | DeepJDOT (Ls + alpha d) · USPS -> MNIST | target accuracy: 95.53 -> 96.4 | Improved |
| apply DeepJDOT adaptation | source-only CNN · MNIST -> MNIST-M | target accuracy: 60.8 -> 92.4 | Improved |
| apply DeepJDOT adaptation | source-only CNN · MNIST -> USPS | target accuracy: 94.8 -> 95.7 | No clear effect |
| apply DeepJDOT adaptation | source-only ResNet-50 · VisDA-2017 mean | accuracy: 28.0 -> 66.9 | Improved |
| apply DeepJDOT adaptation | source-only ResNet-50 · VisDA-2017 car class | accuracy: 94.7 -> 87.3 | Hurt |
| use DeepJDOT instead of DAH | VGG-16 · Office-Home mean | accuracy: 45.54 -> 50.67 | Improved |
Later papers that compared with this method
4 later accepted papers report DeepJDOT as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- Gradual Domain Adaptation: Theory and Algorithms (JMLR 2024 · better on 4 of 4 shared results)
- Towards Unsupervised Domain Adaptation via Domain-Transformer (IJCV 2024 · better on 1 of 1 shared result)
- Reliable Weighted Optimal Transport for Unsupervised Domain Adaptation (CVPR 2020 · better on 1 of 1 shared result)
- Structure-Aware Feature Fusion for Unsupervised Domain Adaptation (AAAI 2020 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Optimal Transport for Domain Adaptation (TPAMI 2016 · similarity 0.89)
- Beyond Sharing Weights for Deep Domain Adaptation (TPAMI 2018 · similarity 0.87)
- Transferable Representation Learning with Deep Adaptation Networks (TPAMI 2018 · similarity 0.87)
- Joint distribution optimal transportation for domain adaptation (NeurIPS 2017 · similarity 0.87)
- Rethinking Joint Maximum Mean Discrepancy for Visual Domain Adaptation (NeurIPS 2025 · similarity 0.86)
- A General Representation-Based Approach to Multi-Source Domain Adaptation (ICML 2025 · similarity 0.85)
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.
