Research map / Optimization / Implicit bias and generalization
Orthogonal SVD Covariance Conditioning and Latent Disentanglement
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 |
|---|---|---|---|
| add Nearest Orthogonal Gradient (NOG) (ablation) | SVD decorrelated BN, ResNet-50 · CIFAR100 | validation error (%): 19.99 -> 19.43 | Improved |
| add Optimal Learning Rate (OLR) (ablation) | SVD decorrelated BN, ResNet-50 · CIFAR10 | validation error (%): 4.35 -> 4.23 | Improved |
| add orthogonal weight (OW) to SVD (ablation) | SVD decorrelated BN, ResNet-50 · CIFAR10 | validation error (%): 4.35 -> 4.42 | Hurt |
| add orthogonal weight and OLR on top of NOG (ablation) | SVD + NOG · CIFAR100 | validation error (%): 19.43 -> 19.05 | Improved |
| add NOG (ablation) | SVD decorrelated BN, DenseNet-121 · CIFAR100 | validation error (%): 27.37 -> 25.14 | Improved |
| add NOG + OW + OLR (ablation) | SVD decorrelated BN, MobileNet-v2 · CIFAR100 | validation error (%): 34.35 -> 32.83 | Improved |
| add orthogonal loss (OL) to SVD (ablation) | GCP ResNet-18 · ImageNet | top-1 accuracy (%): 73.13 -> 71.75 | Hurt |
| add NOG (ablation) | GCP ResNet-18 · ImageNet | top-1 accuracy (%): 73.13 -> 73.51 | Improved |
| add NOG + OW + OLR (ablation) | GCP ResNet-18 · ImageNet | top-1 accuracy (%): 73.13 -> 73.82 | Improved |
| add NOG and OLR (ablation) | SVD solver, GCP ResNet-18 · ImageNet | SVD failure times: 5 -> 0 | Improved |
| add NOG + OLR (ablation) | EigenGAN · AnimeFace | FID: 23.59 -> 16.31 | Improved |
| add NOG + OLR (ablation) | EigenGAN · FFHQ | VP: 31.79 -> 39.32 | Improved |
Closest papers by meaning
- Improving Covariance Conditioning of the SVD Meta-Layer by Orthogonality (ECCV 2022 · similarity 0.96)
- Revitalizing SVD for Global Covariance Pooling: Halley’s Method to Overcome Over-Flattening (NeurIPS 2025 · similarity 0.87)
- Orthogonal Jacobian Regularization for Unsupervised Disentanglement in Image Generation (ICCV 2021 · similarity 0.86)
- Can We Gain More from Orthogonality Regularizations in Training Deep Networks? (NeurIPS 2018 · similarity 0.85)
- Orthogonal Subspace Decomposition for Generalizable AI-Generated Image Detection (ICML 2025 · similarity 0.85)
- Rethinking Invariance Regularization in Adversarial Training to Improve Robustness-Accuracy Trade-off (ICLR 2025 · similarity 0.85)
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.
