Research map / Learning theory and algorithms / Matrix and tensor methods
Fast Differentiable Matrix Square Root and Inverse Square Root
Yue Song, Nicu Sebe, Wei Wang
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 |
|---|---|---|---|
| increase MPA forward degree from [3,3] to [5,5] (ablation) | MPA-Lya · CIFAR10 ZCA whitening (ResNet-18) | validation error (%): 4.64 -> 4.39 | Improved |
| increase MPA forward degree from [5,5] to [6,6] (ablation) | MPA-Lya · CIFAR10 ZCA whitening (ResNet-18) | validation error (%): 4.39 -> 4.45 | No clear effect |
| replace exact Bartels-Stewart backward with iterative Lyapunov solver (8 iterations) (ablation) | MPA-Lya · CIFAR10 ZCA whitening (ResNet-18) | validation error (%): 4.57 -> 4.39 | Improved |
| increase Lyapunov backward iterations from 5 to 8 (ablation) | MPA-Lya · CIFAR10 ZCA whitening (ResNet-18) | validation error (%): 4.48 -> 4.39 | Improved |
| increase Lyapunov backward iterations from 8 to 9 (ablation) | MPA-Lya · CIFAR10 ZCA whitening (ResNet-18) | validation error (%): 4.39 -> 4.40 | No clear effect |
| use Pade approximant (MPA) instead of Taylor polynomial (MTP) in the forward pass | Lyapunov backward · CIFAR100 ZCA whitening, ResNet-50 | validation error (%): 20.55 -> 19.55 | Improved |
| use MPA-Lya instead of NS iteration | ZCA whitening · CIFAR100 ZCA whitening, ResNet-50 | validation error (%): 19.39 -> 19.55 | Hurt |
| use MPA-Lya instead of NS iteration for covariance square root | GCP ResNet-50 · ImageNet | time FP+BP (ms): 164.43 -> 110.61 | Improved |
| use MPA-Lya instead of NS iteration | temporal-attentive GCP · HMDB51 | top-1 accuracy (%): 72.75 -> 74.05 | Improved |
| use MPA-Lya instead of NS iteration | So-ViT-7 · ImageNet | top-1 accuracy (%): 76.50 -> 76.84 | Improved |
| use MPA-Lya instead of SVD-Padé | fine-grained GCP · Cars | accuracy (%): 92.8 -> 92.5 | Hurt |
| use MPA-Lya instead of NS iteration | WCT style transfer · Artworks | LPIPS: 0.5578 -> 0.5615 | Improved |
Closest papers by meaning
- Fast Differentiable Matrix Square Root (ICLR 2022 · similarity 0.97)
- A Sparse Resultant Based Method for Efficient Minimal Solvers (CVPR 2020 · similarity 0.84)
- Robust Differentiable SVD (TPAMI 2021 · similarity 0.84)
- Solving Minimal Problems Without Matrix Inversion Using FFT-Based Interpolation (CVPR 2026 · similarity 0.84)
- Can We Remove the Square-Root in Adaptive Gradient Methods? A Second-Order Perspective (ICML 2024 · similarity 0.81)
- Globally Q-linear Gauss-Newton Method for Overparameterized Non-convex Matrix Sensing (NeurIPS 2024 · similarity 0.81)
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.
