Research map / Label-efficient and robust learning / Domain adaptation and generalization
Domain-Adversarial Training of Neural Networks
Yaroslav Ganin, Evgeniya Ustinova, Hana Ajakan, Pascal Germain, Hugo Larochelle, François Laviolette, Mario March, Victor Lempitsky
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 |
|---|---|---|---|
| add domain-adversarial regularizer to a shallow NN (ablation) | one-hidden-layer NN · Amazon reviews (original data) | Poisson binomial win probability: .87 | Improved |
| use DANN instead of linear SVM | DANN · Amazon reviews (original data) | Poisson binomial win probability: .83 | Improved |
| run DANN on mSDA representations instead of NN | DANN on mSDA · Amazon reviews (mSDA representation) | Poisson binomial win probability: .92 | Improved |
| run DANN on mSDA representations instead of SVM | DANN on mSDA · Amazon reviews (mSDA representation) | Poisson binomial win probability: .88 | Improved |
| add domain-adversarial training to CNN (ablation) | source-only CNN · MNIST to MNIST-M | accuracy: .5225 -> .7666 | Improved |
| add domain-adversarial training to CNN (ablation) | source-only CNN · Syn Numbers to SVHN | accuracy: .8674 -> .9109 | Improved |
| add domain-adversarial training to CNN (ablation) | source-only CNN · SVHN to MNIST | accuracy: .5490 -> .7385 | Improved |
| add domain-adversarial training to CNN (ablation) | source-only CNN · Syn Signs to GTSRB | accuracy: .7900 -> .8865 | Improved |
| replace DANN with subspace alignment | source-only CNN · Syn Numbers to SVHN | accuracy: .8674 -> .8644 | Hurt |
| add domain-adversarial training (ablation) | source-only CNN · MNIST to SVHN | accuracy: approximately 0.25 accuracy | No clear effect |
| add domain-adversarial training to AlexNet (ablation) | source-only AlexNet · Office Amazon to Webcam | accuracy: .642 -> .730 | Improved |
| add domain-adversarial training to AlexNet (ablation) | source-only AlexNet · Office DSLR to Webcam | accuracy: .961 -> .964 | Improved |
Closest papers by meaning
- f-Domain Adversarial Learning: Theory and Algorithms (ICML 2021 · similarity 0.93)
- A DIRT-T Approach to Unsupervised Domain Adaptation (ICLR 2018 · similarity 0.92)
- Transferable Adversarial Training: A General Approach to Adapting Deep Classifiers (ICML 2019 · similarity 0.92)
- Adversarial Multiple Source Domain Adaptation (NeurIPS 2018 · similarity 0.92)
- Adversarial Discriminative Domain Adaptation (CVPR 2017 · similarity 0.91)
- Domain-Symmetric Networks for Adversarial Domain Adaptation (CVPR 2019 · similarity 0.91)
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.
