Research map / Learning theory and algorithms / Kernels and regression theory
Hybrid Random 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 |
|---|---|---|---|
| Use angular hybrid random features rather than trigonometric or positive random features | Softmax kernel estimation on fixed-length inputs · Sphere S(r), r ≥ 1 | max-relative-error: 1/r | Helped |
| Share the same random projections between the trigonometric and positive estimators (ablation) | Bipolar hybrid estimator · Softmax kernel estimation | MSE: -(2/m) SM²(x,y)(1-cos(||x||₂²-||y||₂²)) E[λ̂ₙ(1-λ̂ₙ)] | Helped |
| Use angular rather than Gaussian lambda coefficients (ablation) | Hybrid random features · Fixed-length softmax kernel estimation | variance | Helped |
| Use orthogonal rather than iid random features in HRF (ablation) | HRF softmax kernel estimator · Wine | MSE (10^-3): 0.85 ± 0.05 -> 0.70 ± 0.08 | Helped |
| Use orthogonal rather than iid random features in HRF (ablation) | HRF softmax kernel estimator · Boston | MSE (10^-3): 0.79 ± 0.08 -> 0.72 ± 0.06 | Helped |
| Use clustering-based HRF variants rather than angular HRF | Conformer-Transducer implicit attention · LibriSpeech | WER: 2.03 ± 0.08% -> 1.72 ± 0.02% | Helped |
| Replace FAVOR+ with HRF | Conformer-Transducer implicit attention · LibriSpeech | WER: 2.65 ± 0.06% -> 1.72 ± 0.02% | Helped |
| Replace trigonometric random features with HRF | Conformer-Transducer implicit attention · LibriSpeech | WER: 3.12 ± 0.05% -> 1.72 ± 0.02% | Helped |
| Use HRFs instead of baseline methods | Robotics attention models · Quadruped locomotion and robotic-arm manipulation | FLOPS at same policy quality: 3x fewer FLOPS | Helped |
Closest papers by meaning
- Learning Kernels with Random Features (NeurIPS 2016 · similarity 0.89)
- Simplex Random Features (ICML 2023 · similarity 0.89)
- Manifold Random Features (NeurIPS 2026 · similarity 0.87)
- Random features for Grassmannian kernel approximation with bounded rank-one projections (NeurIPS 2026 · similarity 0.85)
- General Graph Random Features (ICLR 2024 · similarity 0.85)
- Variance-Reducing Couplings for Random Features (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.