Research map / Learning theory and algorithms / Kernels and regression theory
Quantization Algorithms for Random Fourier Features
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 |
|---|---|---|---|
| Replace full-precision RFF with low-bit LM-RFF | Kernel SVM · UCI and ASU datasets | test accuracy | Helped |
| Use 1-bit LM-RFF instead of 1-bit StocQ | Kernel SVM · UCI and ASU datasets | test accuracy | Helped |
| Use 2-bit LM-RFF | Kernel SVM · UCI and ASU datasets | test accuracy | Helped |
| Increase StocQ precision to 4 bits | Kernel SVM · UCI and ASU datasets | test accuracy | No clear effect |
| Replace StocQ with LM-RFF at the same bit width and number of features | Kernel Ridge Regression · Synthetic dataset | test MSE: 14.5 vs. 5.9 (1-bit); 5.0 vs. 4.1 (2-bit) | Helped |
| Use LM-RFF rather than full-precision RFF at constrained memory | Kernel Ridge Regression · Synthetic dataset | test MSE: 3.5 -> 5.9 (1-bit); 3.5 -> 4.1 (2-bit) | Hurt |
| Use 1-bit LM-RFF instead of 2-bit StocQ | Kernel Ridge Regression · Synthetic dataset | test MSE: 14.5 -> 5.9 | Helped |
| Use LM-RFF rather than full-precision RFF to reach 95% accuracy | Kernel SVM · Isolet | memory per sample: ≈ 18000 bits -> ≈ 2000 bits; 9x compression ratio | Helped |
| Use LM-RFF rather than StocQ at comparable accuracy | Kernel SVM · UCI and ASU datasets | compression ratio: 2x∼4x | Helped |
| Use LM-RFF rather than StocQ | Kernel Ridge Regression · Synthetic dataset | test MSE | Helped |
| Use LM-RFF rather than StocQ at equal bit width | Kernel matrix approximation · Isolet and BASEHOCK | scale-invariant kernel approximation error | Helped |
| Add normalization to the LM quantized kernel estimator (ablation) | LM-RFF and LM²-RFF kernel estimators · Kernel estimation, ρ∈[0,1] when γ≤π/√2 | debiased variance | Helped |
Closest papers by meaning
- On The Relative Error of Random Fourier Features for Preserving Kernel Distance (ICLR 2023 · similarity 0.89)
- A random matrix analysis of random Fourier features: beyond the Gaussian kernel, a precise phase transition, and the corresponding double descent (NeurIPS 2020 · similarity 0.88)
- TRF: Learning Kernels with Tuned Random Features (AAAI 2022 · similarity 0.88)
- Scalable Random Wavelet Features: Efficient Non-Stationary Kernel Approximation with Convergence Guarantees (ICLR 2026 · similarity 0.87)
- Random features for Grassmannian kernel approximation with bounded rank-one projections (NeurIPS 2026 · similarity 0.85)
- Fast Summation of Radial Kernels via QMC Slicing (ICLR 2025 · 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.