Research map / Learning theory and algorithms / Matrix and tensor methods
KoPA: Automated Kronecker Product Approximation
Chencheng Cai, Rong Chen, Han Xiao
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 Kronecker product approximation instead of rank-1 SVD at equal parameter count | cameraman image · Cameraman, 1023 parameters | variance explained: 45.63% -> 77.55% | Improved |
| use one-term KoPA instead of one-term SVD for denoising | denoising of 512x512 images · Image denoising, cameraman | reconstruction error: 0.5446 -> 0.1337 | Improved |
| use multi-term KoPA instead of multi-term SVD for denoising | denoising of 512x512 images · Image denoising, cameraman | reconstruction error: 0.0644 -> 0.0399 | Improved |
| use total variation regularization instead of multi-term KoPA | denoising of 512x512 images · Image denoising, cameraman | reconstruction error: 0.0399 -> 0.0294 | Improved |
| add more Kronecker terms (multi-term vs one-term KoPA) | KoPA · Image denoising, mandril | reconstruction error: 0.3557 -> 0.0889 | Improved |
| use KoPA over low rank approximation across noise levels sigma 0.1, 0.2, 0.3 with BIC-selected configuration | multi-term KoPA · Noisy cameraman | reconstruction error vs number of parameters | Improved |
| increase the stronger penalty from AIC to BIC for configuration selection (ablation) | KoPA configuration selection · Cameraman image | selected configuration: (8,9) -> (6,7) | Mixed |
| use MSE (kappa=0) instead of AIC (kappa=2) for configuration selection (ablation) | KoPA configuration selection · Simulation, one-term model | correct selection frequency | Hurt |
| increase representation gap psi^2 (ablation) | KoPA configuration selection · Simulation, one-term model | correct selection frequency | Improved |
| increase the second-term strength ratio lambda2^2/lambda1^2 (ablation) | MSE selection in two-term model · Simulation, two-term model | correct selection frequency | Hurt |
Closest papers by meaning
- Subquadratic Kronecker Regression with Applications to Tensor Decomposition (NeurIPS 2022 · similarity 0.86)
- Robust Kronecker-Decomposable Component Analysis for Low-Rank Modeling (ICCV 2017 · similarity 0.85)
- A Trace-restricted Kronecker-Factored Approximation to Natural Gradient (AAAI 2021 · similarity 0.85)
- Fast Inference with Kronecker-Sparse Matrices (ICML 2025 · similarity 0.85)
- A Kronecker-factored approximate Fisher matrix for convolution layers (ICML 2016 · similarity 0.85)
- Matrix Denoising with Doubly Heteroscedastic Noise: Fundamental Limits and Optimal Spectral Methods (NeurIPS 2024 · similarity 0.84)
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.
