Research map / Learning theory and algorithms / Kernels and regression theory
A Kernel Theory of Modern Data Augmentation
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 |
|---|---|---|---|
| Apply rotation augmentation to training data | RBF kernel classifier with 10,000 random Fourier features · MNIST | accuracy (%): 96.1 $\pm$ 0.1 -> 97.6 $\pm$ 0.2 | Helped |
| Apply rotation augmentation directly to random Fourier features instead of augmenting raw data | RBF kernel classifier with 10,000 random Fourier features · MNIST | training time: 220s -> 54s | Helped |
| Apply rotation augmentation directly to random Fourier features instead of augmenting raw data | RBF kernel classifier with 10,000 random Fourier features · MNIST | accuracy (%): 97.6 $\pm$ 0.2 -> 97.6 $\pm$ 0.1 | No clear effect |
| Apply rotation augmentation to training data | RBF kernel classifier with 10,000 random Fourier features · CIFAR-10 | accuracy (%): 39.4 $\pm$ 0.5 -> 45.3 $\pm$ 0.5 | Helped |
| Apply rotation augmentation directly to random Fourier features instead of augmenting raw data | RBF kernel classifier with 10,000 random Fourier features · CIFAR-10 | training time: 291s -> 124s | Helped |
| Apply rotation augmentation directly to random Fourier features instead of augmenting raw data | RBF kernel classifier with 10,000 random Fourier features · CIFAR-10 | accuracy (%): 45.3 $\pm$ 0.5 -> 45.2 $\pm$ 0.4 | No clear effect |
| Apply rotation augmentation to training data | RBF kernel classifier with 10,000 random Fourier features · DDSM | accuracy (%): 57.3 $\pm$ 6.7 -> 59.4 $\pm$ 3.2 | No clear effect |
| Apply rotation augmentation directly to random Fourier features instead of augmenting raw data | RBF kernel classifier with 10,000 random Fourier features · DDSM | training time: 61s -> 34s | Helped |
| Apply rotation augmentation directly to random Fourier features instead of augmenting raw data | RBF kernel classifier with 10,000 random Fourier features · DDSM | accuracy (%): 59.4 $\pm$ 3.2 -> 58.8 $\pm$ 5.1 | No clear effect |
| Include the second-order variance regularization term in addition to feature averaging | Augmented-objective approximation · MNIST and CIFAR-10 | objective approximation quality | Helped |
| Use first- and second-order terms rather than only the second-order term | Tangent propagation approximation · Figure experiment | test prediction disagreement: 6--9x reduction | Helped |
| Average transformed features at an intermediate network layer rather than applying approximation before the first layer | LeNet · MNIST and CIFAR-10 | accuracy gain retained: 30%; 92% | Helped |
Closest papers by meaning
- Data Augmentation as Feature Manipulation (ICML 2022 · similarity 0.90)
- Optimizing Data Augmentation through Bayesian Model Selection (ICLR 2026 · similarity 0.89)
- Characterizing the Generalization Error of Random Feature Regression with Arbitrary Data-Augmentation (NeurIPS 2026 · similarity 0.87)
- Understanding Augmentation-based Self-Supervised Representation Learning via RKHS Approximation and Regression (ICLR 2024 · similarity 0.87)
- First-Order Manifold Data Augmentation for Regression Learning (ICML 2024 · similarity 0.86)
- Data Augmentation with Diffusion for Open-Set Semi-Supervised Learning (NeurIPS 2024 · 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.