Research map / Learning theory and algorithms / Kernels and regression theory
Chefs' Random Tables: Non-Trigonometric 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 |
|---|---|---|---|
| Replace FAVOR+ attention features with FAVOR++ | low-rank Transformers · GLUE Dev MNLI | accuracy: 80.26 -> 81.25 | Helped |
| Replace FAVOR+ attention features with FAVOR++ | low-rank Transformers · GLUE Dev QQP | accuracy: 89.53 -> 90.15 | Helped |
| Replace FAVOR+ attention features with FAVOR++ | low-rank Transformers · GLUE Dev QNLI | accuracy: 87.13 -> 89.58 | Helped |
| Replace FAVOR+ attention features with FAVOR++ | low-rank Transformers · GLUE Dev SST-2 | accuracy: 90.58 -> 92.00 | Helped |
| Replace FAVOR+ attention features with FAVOR++ | low-rank Transformers · GLUE Dev CoLA | MCC: 53.17 -> 54.95 | Helped |
| Replace FAVOR+ attention features with FAVOR++ | low-rank Transformers · GLUE Dev STS-B | Spearman correlation: 85.07 -> 85.62 | Helped |
| Replace FAVOR+ attention features with FAVOR++ | low-rank Transformers · GLUE Dev MRPC | F1: 83.82 -> 85.78 | Helped |
| Replace FAVOR+ attention features with FAVOR++ | low-rank Transformers · GLUE Dev RTE | accuracy: 67.59 -> 67.87 | Helped |
| Use OPRF instead of TrigRF features | Gaussian-kernel non-parametric classification · UCI benchmark average | test accuracy (%): 35.5 -> 57.8 | Helped |
| Warm-start FAVOR++ from a pre-trained BERT-base checkpoint (ablation) | FAVOR++ · GLUE Dev | task scores: 81.25 -> 82.29; 90.15 -> 90.43; 89.58 -> 89.73; 92.00 -> 92.20; 54.95 -> 58.85; 85.62 -> 85.90; 85.78 -> 88.73; 67.87 -> | Mixed |
| Use orthogonal random projections with OPRFs instead of i.i.d. projections | OPRF kernel estimator · Gaussian-kernel estimation | variance | Helped |
| Use bounded positive OPRFs rather than unbounded positive features | softmax-kernel estimation · Theoretical concentration | tail bound | Helped |
Closest papers by meaning
- Simplex Random Features (ICML 2023 · similarity 0.89)
- Dense-Exponential Random Features: Sharp Positive Estimators of the Gaussian Kernel (NeurIPS 2023 · similarity 0.89)
- 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)
- On the Minimal Degree Bias in Generalization on the Unseen for non-Boolean Functions (ICML 2024 · similarity 0.85)
- Manifold Random Features (NeurIPS 2026 · 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.