Research map / Learning theory and algorithms / Kernels and regression theory
An Empirical Study on The Properties of Random Bases for Kernel Methods
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 |
|---|---|---|---|
| unsupervised kernel-approximation-error adaptation of random basis (UAB) | Gaussian and ArcCos random features · Gisette, MNIST, CoverType, CIFAR10 | accuracy and KAE | Mixed |
| supervised kernel target alignment adaptation (SAB) | Gaussian kernel features · CIFAR10 | accuracy: 76.4 -> 79.0 | Helped |
| supervised kernel target alignment adaptation (SAB) | ArcCos kernel features · CoverType | accuracy: 83.6 -> 88.7 is an improvement; SAB below DAB (88.7 vs 92.9) | Hurt |
| supervised kernel target alignment adaptation (SAB) | Gaussian kernel features · CoverType | accuracy: 91.9 -> 90.4 | Hurt |
| transfer of task-adapted basis from MNIST 5-class task to remaining classes | SAB vs RB · MNIST transfer task | accuracy | Helped |
| transfer of discriminatively adapted basis (DAB) to new task | DAB · MNIST transfer task | accuracy | Hurt |
| stacking ArcCos layers (ArcCos2/ArcCos3) with the same feature count | ArcCos deep kernels · CoverType | accuracy | Mixed |
| replacing Gaussian matrix with structured Hadamard blocks (fast kernel machine) with one block | random basis (RB) Gaussian kernel · MNIST and CoverType | accuracy | Hurt |
| unsupervised adaptation of structured Hadamard basis | UAB fast kernel machine · MNIST and CoverType | accuracy | Helped |
| supervised adaptation of structured Hadamard basis | SAB fast kernel machine · MNIST and CoverType | accuracy | No clear effect |
| discriminative adaptation with structured Hadamard blocks | DAB fast kernel machine · MNIST and CoverType | accuracy | Hurt |
| adapting the basis (UAB, SAB, DAB) versus random basis on Gisette and MNIST with Gaussian kernel | Gaussian kernel features · Gisette | accuracy: RB 98.1 vs UAB 97.9, SAB 98.1, DAB 97.9 | No clear effect |
Closest papers by meaning
- Learning Kernels with Random Features (NeurIPS 2016 · similarity 0.90)
- Adaptive kernel predictors from feature-learning infinite limits of neural networks (ICML 2025 · similarity 0.87)
- Toward Deeper Understanding of Neural Networks: The Power of Initialization and a Dual View on Expressivity (NeurIPS 2016 · similarity 0.87)
- The Comparative Power of ReLU Networks and Polynomial Kernels in the Presence of Sparse Latent Structure (ICLR 2019 · similarity 0.87)
- From Kernels to Features: A Multi-Scale Adaptive Theory of Feature Learning (ICML 2025 · similarity 0.87)
- Faithful and Efficient Explanations for Neural Networks via Neural Tangent Kernel Surrogate Models (ICLR 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.