Research map / Learning theory and algorithms / Kernels and regression theory
Random Feature Maps for the Itemset Kernel
Kyohei Atarashi, Subhransu Maji, Satoshi Oyama
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 |
|---|---|---|---|
| Rademacher instead of Laplace random vectors (ablation) | RK feature map · Movielens 100K, second-order ANOVA, D = 2d | absolute error: 8.29e-4 -> 6.53e-4 | Improved |
| Rademacher instead of Gaussian random vectors (ablation) | RK feature map · Movielens 100K, third-order ANOVA, D = 2d | absolute error: 2.67e-5 -> 2.26e-5 | Improved |
| different random vector distribution (ablation) | RK feature map · Movielens 100K, all-subsets kernel | absolute error: 4.24e-2 vs 4.15e-2 at D = 2d | No clear effect |
| increase random feature dimension from 2d to 16d | RK (Rademacher) · Movielens 100K, second-order ANOVA | absolute error: 6.53e-4 -> 2.33e-4 | Improved |
| increase random feature dimension from 2d to 16d | RK (Rademacher) · Movielens 100K, all-subsets kernel | absolute error: 4.24e-2 -> 1.49e-2 | Improved |
| signed circulant random feature (SCRK) instead of RK | RK feature map · Movielens 100K, third-order ANOVA | absolute error: 2.26e-5 -> 2.29e-5 | No clear effect |
| SCRK instead of RK (even order) | RK feature map · Movielens 100K, second-order ANOVA, D = 2d | absolute error: 6.53e-4 -> 7.22e-4 | Hurt |
| SCRK instead of RK at large original dimension d | RK feature map · synthetic data, d up to 4096 | mapping time | Improved |
| SCRK instead of RK at d = 512 | RK feature map · synthetic data | mapping time | Hurt |
| linear SVM with RK map (D = 1,248) instead of nonlinear SVM / FM | linear SVM · Movielens 100K binarized | training and testing time: 2-5 times less | Improved |
| linear SVM with RK map (D = 1,248) instead of nonlinear SVM / FM | linear SVM · Movielens 100K binarized | accuracy | No clear effect |
| RM map for polynomial kernel instead of RK map | linear SVM · Movielens 100K binarized | accuracy | Improved |
Closest papers by meaning
- Learning Kernels with Random Features (NeurIPS 2016 · similarity 0.86)
- Quadrature-based features for kernel approximation (NeurIPS 2018 · similarity 0.86)
- Random Features for Kernel Approximation: A Survey on Algorithms, Theory, and Beyond (TPAMI 2021 · similarity 0.86)
- Random features for Grassmannian kernel approximation with bounded rank-one projections (NeurIPS 2026 · similarity 0.85)
- On the Efficiency of ERM in Feature Learning (NeurIPS 2024 · similarity 0.84)
- Supervised Kernel Thinning (NeurIPS 2024 · similarity 0.83)
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.
