Research map / Label-efficient and robust learning / Domain adaptation and generalization
Generalizing Across Domains via Cross-Gradient Training
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 domain-guided cross-gradient augmentation instead of standard training | Classifier · Fonts, Handwriting, MNIST, Speech | accuracy: 68.5 -> 72.6; 82.5 -> 88.6; 95.6 -> 98.6; 72.6 -> 73.5 | Helped |
| Use domain-guided perturbations instead of label-loss perturbations | CrossGrad classifier · Fonts, Handwriting, MNIST, Speech | accuracy: 71.4 -> 72.6; 86.3 -> 88.6; 97.8 -> 98.6; 72.7 -> 73.5 | Helped |
| Use domain-adversarial training instead of standard training | Classifier · Fonts, Handwriting, MNIST, Speech | accuracy: 68.5 -> 68.9; 82.5 -> 83.8; 95.6 -> 98.0; 72.6 -> 70.4 | Mixed |
| Use CrossGrad with a 2-block ResNet instead of standard training | Classifier · Fonts and Handwriting | accuracy: 80.2 -> 82.4; 91.5 -> 92.1 | Helped |
| Increase the number of available training domains | CrossGrad · Speech | accuracy gain over baseline: +2.3; +0.9; +0.7; -0.4 | Mixed |
| Use CrossGrad for an intermediate held-out rotation domain rather than a non-CrossGrad comparator | Rotated MNIST classifier · M15, M30, M45, M60 | accuracy: M15: 98.0 -> 98.6; M30: 97.8 -> 98.0; M45: 97.4 -> 97.7; M60: 96.9 -> 97.7 | Mixed |
| Use CrossGrad for an extreme held-out rotation domain rather than the strongest listed comparator | Rotated MNIST classifier · M0 and M75 | accuracy: 89.7 -> 88.3; 92.1 -> 91.4 | Hurt |
Closest papers by meaning
- Adaptive Methods for Real-World Domain Generalization (CVPR 2021 · similarity 0.90)
- Learning to Generalize across Domains on Single Test Samples (ICLR 2022 · similarity 0.90)
- Domain Generalization via Entropy Regularization (NeurIPS 2020 · similarity 0.90)
- Cross-Domain Gradient Discrepancy Minimization for Unsupervised Domain Adaptation (CVPR 2021 · similarity 0.90)
- Feature-Critic Networks for Heterogeneous Domain Generalization (ICML 2019 · similarity 0.90)
- A General Representation-Based Approach to Multi-Source Domain Adaptation (ICML 2025 · 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.