Research map / Learning theory and algorithms / Matrix and tensor methods
A generalization of the randomized singular value decomposition
Nicolas Boulle, Alex Townsend
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 the discretized negative-Laplacian Green's function as the input covariance | Randomized SVD for the inverse differential-operator matrix · 2000×2000 inverse differential-operator matrix | approximation error: 1.3-$1.6 | Improved |
| Use the discretized negative-Laplacian Green's function as the input covariance | Randomized SVD for the inverse differential-operator matrix · 2000×2000 inverse differential-operator matrix | computational time: 20% | Hurt |
| Use a Jacobi covariance kernel with eigenvalues 1/j³ instead of squared-exponential covariance with ℓ=0.01 (ablation) | Randomized SVD for HS operators · HS operator approximation, k=100 | relative approximation error: 2.6\times 10^{-11} vs. 5.7\times 10^{-13}; ratio 45.6 | Hurt |
| Use covariance kernels with slower eigenvalue decay (ablation) | Randomized SVD for HS operators · HS kernels with high numerical rank | relative approximation error | Improved |
| Use squared-exponential covariance with ℓ=1 or ℓ=0.1 (ablation) | Randomized SVD for HS operators · HS kernels with high numerical rank | approximation ability: j=13; j=59 | Hurt |
| Use the Jacobi kernel with eigenvalues 1/j³ rather than squared-exponential covariance with ℓ=0.01 (ablation) | Randomized SVD for HS operators · HS operator approximation, k=91 | covariance quality factor γ₉₁: 5.88\times 10^{-2} vs. 4.24\times 10^{-6} | Hurt |
Closest papers by meaning
- Fast Randomized Singular Value Thresholding for Low-Rank Optimization (TPAMI 2017 · similarity 0.87)
- Operator SVD with Neural Networks via Nested Low-Rank Approximation (ICML 2024 · similarity 0.85)
- Entrywise error bounds for low-rank approximations of kernel matrices (NeurIPS 2024 · similarity 0.84)
- Integration-free Kernels for Equivariant Gaussian Process Modelling (ICML 2025 · similarity 0.83)
- Sparse Gaussian Processes: Structured Approximations and Power-EP Revisited (NeurIPS 2025 · similarity 0.83)
- Fast Updating Truncated SVD for Representation Learning with Sparse Matrices (ICLR 2024 · similarity 0.83)
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.
