Research map / Learning theory and algorithms / Kernels and regression theory
Alternating Circulant Random Features for Semigroup Kernels
Yusuke Mukuta, Yoshitaka Ushiku, Tatsuya Harada
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 |
|---|---|---|---|
| remove random sign flipping (circ) instead of alternating mixing (ablation) | altcirc · CUB BoVW, D=d | accuracy (%): 11.26 -> 9.37 | Hurt |
| remove random sign flipping (circ) instead of alternating mixing (ablation) | altcirc · Stanford Dogs BoVW, D=d | accuracy (%): 16.83 -> 12.50 | Hurt |
| remove random sign flipping (circ) instead of alternating mixing (ablation) | altcirc · Caltech256 VGG-16 softmax, D=d | accuracy (%): 69.78 -> 67.35 | Hurt |
| increase mixed matrices from m = 2 to m = log2 d (ablation) | altcirc · CUB VGG-16 softmax, D=d | accuracy (%): 33.76 -> 36.05 | Improved |
| increase mixed matrices from m = 2 to m = log2 d (ablation) | altcirc · Caltech256 VGG-16 softmax, D=d | accuracy (%): 68.54 -> 69.78 | Improved |
| increase mixed matrices from m = 2 to m = log2 d (ablation) | altcirc · Caltech256 BoVW, D=16d | accuracy (%): 36.30 -> 36.09 | No clear effect |
| replace random Laplace features with altcirc (m = log2 d) | semigroup kernel features · CUB VGG-16 softmax, D=16d | accuracy (%): 42.55 -> 42.46 | No clear effect |
| replace Gaussian kernel with exponential-semigroup kernel (random Laplace) | kernel choice · CUB VGG-16 softmax, D=16d | accuracy (%): 23.44 -> 42.55 | Improved |
| replace Gaussian kernel with exponential-semigroup kernel (random Laplace) | kernel choice · Caltech256 VGG-16 softmax, D=16d | accuracy (%): 58.69 -> 72.74 | Improved |
| replace Gaussian kernel with semigroup kernel on last-activation features | kernel choice · CUB VGG-16 last activation, D=d | accuracy (%): 56.88 -> 55.00 | Hurt |
| replace Gaussian kernel with altcirc (m = 2) as dimension grows | kernel choice · CUB VGG-16 last activation, D=16d | accuracy (%): 57.87 -> 58.94 | Improved |
| increase output dimension from D = d to D = 16d | altcirc (m = log2 d) · Caltech256 BoVW | accuracy (%): 30.06 -> 36.09 | Improved |
Closest papers by meaning
- Randomized Kernel Selection With Spectra of Multilevel Circulant Matrices (AAAI 2018 · similarity 0.87)
- Orthogonal Random Features (NeurIPS 2016 · similarity 0.86)
- Random Fourier Features for Kernel Ridge Regression: Approximation Bounds and Statistical Guarantees (ICML 2017 · similarity 0.85)
- A random matrix analysis of random Fourier features: beyond the Gaussian kernel, a precise phase transition, and the corresponding double descent (NeurIPS 2020 · similarity 0.85)
- Random features for Grassmannian kernel approximation with bounded rank-one projections (NeurIPS 2026 · similarity 0.82)
- Manifold Random Features (NeurIPS 2026 · similarity 0.81)
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.
