Research map / Label-efficient and robust learning / Domain adaptation and generalization
Adversarial Dropout Regularization
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 |
|---|---|---|---|
| Use ADR instead of entropy-based adversarial alignment | domain adaptation for digits · SVHN to MNIST, USPS to MNIST, and MNIST to USPS | accuracy: 93.7$\pm$2.05 -> 94.1$\pm$1.37; 89.2$\pm$3.07 -> 91.5$\pm$3.61; 89.6$\pm$1.44 -> 91.3$\pm$0.65 | Helped |
| Use ADR instead of MMD distribution alignment | ResNet101 fine-tuning · VisDA2017 classification | mean accuracy: 61.1 -> 73.9 | Helped |
| Use ADR instead of MMD distribution alignment | ResNeXt fine-tuning · VisDA2017 classification | mean accuracy: 63.7 -> 79.1 | Helped |
| Retrain a classifier using ADR-generated features (ablation) | ADR with ResNet101 · VisDA2017 classification | mean accuracy: 73.9 -> 74.8 | Helped |
| Retrain a classifier using ADR-generated features (ablation) | ADR with ResNeXt · VisDA2017 classification | mean accuracy: 79.1 -> 79.6 | Helped |
| Use ADR instead of FCN Wild | semantic segmentation · GTA5 to Cityscapes | mIoU: 27.1 -> 33.3 | Helped |
| Use ADR rather than the ALI semi-supervised GAN approach | semi-supervised learning · SVHN | error rate: 7.42$\pm$0.65 -> 6.26$\pm$1.05 | Helped |
| Use ADR rather than the ALI semi-supervised GAN approach | semi-supervised learning · CIFAR10 | error rate: {\bf 17.99}$\pm$1.62 -> 19.63$\pm$0.37 | Hurt |
| Adapt features using ADR rather than leaving the source model unadapted | two-moons classifier · Two moons, target rotated 30 degrees | accuracy: 84\% -> 96\% | Helped |
| Replace the ADR sensitivity loss with classifier-output entropy (ablation) | ENT domain-adaptation baseline · GTA5 to Cityscapes | mIoU: 25.3 -> 15.0 | Hurt |
Closest papers by meaning
- f-Domain Adversarial Learning: Theory and Algorithms (ICML 2021 · similarity 0.89)
- Maximum Classifier Discrepancy for Unsupervised Domain Adaptation (CVPR 2018 · similarity 0.89)
- An Adversarial Perturbation Oriented Domain Adaptation Approach for Semantic Segmentation (AAAI 2020 · similarity 0.89)
- Adversarial Discriminative Domain Adaptation (CVPR 2017 · similarity 0.88)
- Generalizing to Unseen Domains via Adversarial Data Augmentation (NeurIPS 2018 · similarity 0.88)
- Adversarial Dropout for Supervised and Semi-Supervised Learning (AAAI 2018 · similarity 0.88)
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.