Research map / Architectures and efficiency / RNNs and normalization
Batch-Efficient EigenDecomposition for Small and Medium Matrices
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 |
|---|---|---|---|
| use Batched ED instead of torch.svd in decorrelated BN (group 16) | ResNet-18 ZCA whitening · CIFAR10 | validation error: 4.52 -> 4.37 | Improved |
| use Batched ED instead of SVD in decorrelated BN (group 8) | ResNet-18 ZCA whitening · CIFAR100 | validation error: 21.32 -> 20.97 | Improved |
| use Batched ED instead of SVD in decorrelated BN (group 16) | ResNet-18 ZCA whitening · CIFAR100 | validation error: 21.24 -> 21.25 | No clear effect |
| use Batched ED instead of SVD, forward time | ResNet-18 ZCA whitening · CIFAR (group 16) | time (s): 0.172 -> 0.006 | Improved |
| use Batched ED instead of SVD in So-ViT-7 (32x32) | So-ViT · ImageNet | top-1 accuracy: 76.01 -> 76.04 | No clear effect |
| use Batched ED instead of SVD in So-ViT-10 (32x32) | So-ViT · ImageNet | top-1 accuracy: 77.97 -> 77.91 | No clear effect |
| use Batched ED instead of SVD, forward time in So-ViT | So-ViT · ImageNet 768x32x32 | time (s): 0.767 -> 0.431 | Improved |
| use Batched ED instead of SVD in style transfer (group 16) | WCT style transfer · Artworks | LPIPS: 0.5614 -> 0.5694 | Improved |
| use Batched ED instead of SVD in style transfer (group 32) | WCT style transfer · Artworks | LPIPS: 0.5722 -> 0.5700 | No clear effect |
| use Batched ED instead of SVD, forward time in style transfer (group 64) | WCT style transfer · Artworks | time (s): 3.146 -> 0.089 | Improved |
| increase the number of groups in WCT | WCT style transfer · Artworks | visual style strength | Improved |
| use Batched ED instead of SVD for matrices of dimension 32 at small batch | ED solver · numerical test | speed | Hurt |
Closest papers by meaning
- Amortized Eigendecomposition for Neural Networks (NeurIPS 2024 · similarity 0.87)
- Binary Quadratic Quantization: Beyond First-Order Quantization for Real-Valued Matrix Compression (NeurIPS 2025 · similarity 0.83)
- GPLQ: A General, Practical, and Lightning QAT Method for Vision Transformers (NeurIPS 2025 · similarity 0.82)
- FIMA-Q: Post-Training Quantization for Vision Transformers by Fisher Information Matrix Approximation (CVPR 2025 · similarity 0.81)
- Batch Normalization Alleviates the Spectral Bias in Coordinate Networks (CVPR 2024 · similarity 0.81)
- EigenGS Representation: From Eigenspace to Gaussian Image Space (CVPR 2025 · 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.
