Research map / Probabilistic and scientific modeling / Differential equations and PDEs
Deep Learning for Subspace Regression
Vladimir Fanaskov, Vladislav Trifonov, Alexander Rudikov, Ekaterina Muravleva, Ivan Oseledets
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 |
|---|---|---|---|
| predict larger subspace than required (embedding, dim 10 -> 40) (ablation) | subspace regression with L2 loss · D=2 elliptic eigenproblem, first 10 eigenvectors | test error (%): 30% -> 2% | Improved |
| predict larger subspace (embedding) (ablation) | subspace regression · D=2 elliptic eigenproblem | generalisation gap | Improved |
| use subspace regression with L1 loss instead of classical Riemannian interpolation | eigenspace prediction · QM D=2 (a) | relative error (%): 31.9% -> 0.65% | Improved |
| use subspace loss instead of Z2-adjusted l2 eigenvector regression | eigenspace prediction · QM D=1 | relative error (%): 2.33% -> 0.09% | Improved |
| use QR-free stochastic loss L2 instead of L1 as subspace size grows | subspace regression · D=2 training time | training time | Improved |
| use unstabilised L2 (normal equation) loss instead of L1 (ablation) | subspace regression · D=3 elliptic eigenproblem | relative error (%): 24.77% -> 31.46% (size 6); fails at size 24 | Hurt |
| stabilise L2 with Cholesky-QR2 (ablation) | subspace regression · D=3 elliptic eigenproblem | relative error (%): 31.46% -> 28.28% (size 6); 16.3% -> 7.4% (size 48) | Improved |
| initialise LOBPCG with predicted subspace | LOBPCG · parametric eigenproblems | convergence speed / relative error: 2 to 3 times faster; 2 orders lower error | Improved |
| train separate networks per eigenvector with Z2 l2 loss for higher-index eigenvectors | eigenvector regression · D=2 elliptic eigenproblem | test error | Hurt |
Closest papers by meaning
- Optimal deep learning of holomorphic operators between Banach spaces (NeurIPS 2024 · similarity 0.86)
- Neural Networks Perform Sufficient Dimension Reduction (AAAI 2025 · similarity 0.84)
- Latent Neural Operator for Solving Forward and Inverse PDE Problems (NeurIPS 2024 · similarity 0.84)
- Does SGD really happen in tiny subspaces? (ICLR 2025 · similarity 0.84)
- Sharp Generalization for Nonparametric Regression by Over-Parameterized Neural Networks: A Distribution-Free Analysis in Spherical Covariate (ICML 2025 · similarity 0.83)
- Memory-Efficient LLM Training with Online Subspace Descent (NeurIPS 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.
