Research map / Learning theory and algorithms / Kernels and regression theory
Quasi-Monte Carlo Feature Maps for Shift-Invariant Kernels
Haim Avron, Vikas Sindhwani, Jiyan Yang, Michael W. Mahoney
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 |
|---|---|---|---|
| use Halton QMC sequence instead of Monte Carlo random features | random Fourier features Gaussian kernel · census, s=400 | regression error: 0.0791 -> 0.0529 | Improved |
| use Halton QMC sequence instead of Monte Carlo random features | random Fourier features Gaussian kernel · cpu, s=100 | regression error: 0.0383 -> 0.0367 | Improved |
| use Sobol' sequence instead of Monte Carlo | random Fourier features Gaussian kernel · Gram matrix approximation | relative Gram error | No clear effect |
| use scrambled QMC sequences instead of non-scrambled (ablation) | QMC feature maps · Gram matrix approximation | relative Gram error ratio | Improved |
| use QMC sequences instead of MC for classification | approximate kernel ridge regression · classification datasets, s<2000 | accuracy | No clear effect |
| measure box discrepancy on central box b/2 instead of full box (ablation) | box discrepancy as predictor · Gram matrix approximation | correlation with Gram error | Improved |
| add adaptive (learned) QMC sequences over Halton | Halton sequence · cpu and census discrepancy | box discrepancy: 3.41e-3 -> 1.29e-6 | Improved |
| learn only weights on Halton features (Weighted) instead of Halton | Halton sequence · census, s=400, D^b | box discrepancy: 2.61e-3 -> 8.83e-4 | Improved |
| use Global adaptive sequence optimized on full box instead of Halton | kernel ridge regression · census, s=400 | regression error: 0.0529 -> 0.1034 | Hurt |
| use Greedy adaptive (b/4) instead of Halton | kernel ridge regression · cpu, s=500 | regression error: 0.0348 -> 0.0291 | Improved |
| optimize adaptive sequence for the central box b/2 instead of the full box (ablation) | Global adaptive on housing · housing Gram matrix error | mean and norm error | Mixed |
| run Global adaptive optimization for more iterations on housing (ablation) | Global adaptive · housing | Gram error metrics | Hurt |
Closest papers by meaning
- Quasi-Monte Carlo Features for Kernel Approximation (ICML 2024 · similarity 0.91)
- Fast Summation of Radial Kernels via QMC Slicing (ICLR 2025 · similarity 0.87)
- The Polynomial Stein Discrepancy for Assessing Moment Convergence (ICML 2025 · similarity 0.85)
- Variance-Reducing Couplings for Random Features (ICLR 2025 · similarity 0.85)
- Random features for Grassmannian kernel approximation with bounded rank-one projections (NeurIPS 2026 · similarity 0.85)
- Quasi-Monte Carlo for 3D Sliced Wasserstein (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.
