Research map / Learning theory and algorithms / Matrix and tensor methods
What if Neural Networks had SVDs?
Alexander Mathiasen, Frederik Hvilshøj, Jakob Rødsgaard Jørgensen, Anshul Nasery, Davide Mottin
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 |
|---|---|---|---|
| Group Householder matrices into WY products to reduce sequential operations | Orthogonal gradient descent with Householder parameterization · Orthogonal gradient descent | running time: up to 27.1× faster than the sequential algorithm | Improved |
| Use FastH for SVD-reparameterized matrix operations | SVD reparameterization · Matrix inversion, d=768 | speedup over standard inverse: 2.7× | Improved |
| Use FastH for SVD-reparameterized matrix operations | SVD reparameterization · Cayley map, matrix exponential, and determinant, d=768 | speedup over standard method: 3.1×; 4.1×; 3.5× | Improved |
| Use the sequential Householder algorithm instead of FastH | SVD reparameterization · Matrix operations | running time | Hurt |
| Use FastH rather than the prior parallel Householder algorithm | Orthogonal gradient descent · Orthogonal gradient descent | relative speedup: up to 6.2× | Improved |
Closest papers by meaning
- Operator SVD with Neural Networks via Nested Low-Rank Approximation (ICML 2024 · similarity 0.79)
- Faster Algorithms for Structured Linear and Kernel Support Vector Machines (ICLR 2025 · similarity 0.77)
- Sliding Down the Stairs: How Correlated Latent Variables Accelerate Learning with Neural Networks (ICML 2024 · similarity 0.77)
- Searching for Efficient Linear Layers over a Continuous Space of Structured Matrices (NeurIPS 2024 · similarity 0.76)
- Accelerating Data Generation for Neural Operators via Krylov Subspace Recycling (ICLR 2024 · similarity 0.76)
- Feature emergence via margin maximization: case studies in algebraic tasks (ICLR 2024 · similarity 0.76)
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.
