Research map / Architectures and efficiency / CNN design
Can We Gain More from Orthogonality Regularizations in Training Deep Networks?
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 SRIP regularization | Wide ResNet 28-10 · CIFAR-10 | top-1 error rate: 4.16* -> 3.60 | Helped |
| Add SRIP regularization | Wide ResNet 28-10 · CIFAR-100 | top-1 error rate: 20.50* -> 18.19 | Helped |
| Add SRIP regularization | ResNet 34 · ImageNet ILSVRC-2012 val | top-5 error rate: 9.84 -> 8.32 | Helped |
| Add SRIP regularization | Wide ResNet 16-8 · SVHN | error rate: 1.63% -> 1.56% | Helped |
| Add SRIP instead of spectral norm regularization | Wide ResNet 28-10 · CIFAR-10 | top-1 error rate: 3.93% -> 3.60% | Helped |
| Add SRIP instead of OMDSM | Wide ResNet 28-10 · CIFAR-100 | top-1 error rate: 18.76% -> 18.19% | Helped |
| Remove the scheme change and keep SRIP active without the switching schedule (ablation) | Wide ResNet 28-10 with SRIP · CIFAR-10 | top-1 error rate: +0.33% | Hurt |
| Remove the scheme change and keep SRIP active without the switching schedule (ablation) | Wide ResNet 28-10 with SRIP · CIFAR-100 | top-1 error rate: +0.90% | Hurt |
| Use orthogonality regularization and L2 weight decay together throughout training (ablation) | CNNs with orthogonality regularization · CIFAR-10 and CIFAR-100 | top-1 accuracy | Hurt |
| Replace double soft orthogonality with selective soft orthogonality (ablation) | CNNs using DSO · CIFAR-10 and CIFAR-100 | top-1 accuracy | Hurt |
| Add DSO rather than leaving the model unregularized (ablation) | CNNs · CIFAR-10 and CIFAR-100 | top-1 error rate | Mixed |
| Add mutual coherence regularization (ablation) | CNNs · CIFAR-10 and CIFAR-100 | top-1 error rate | Helped |
Closest papers by meaning
- Scaling-up Diverse Orthogonal Convolutional Networks by a Paraunitary Framework (ICML 2022 · similarity 0.88)
- Regularizing CNNs with Locally Constrained Decorrelations (ICLR 2017 · similarity 0.87)
- Regularizing Deep Networks Using Efficient Layerwise Adversarial Training (AAAI 2018 · similarity 0.87)
- An Adaptive Orthogonal Convolution Scheme for Efficient and Flexible CNN Architectures (ICML 2025 · similarity 0.86)
- Revisiting Residual Connections: Orthogonal Updates for Stable and Efficient Deep Networks (NeurIPS 2025 · similarity 0.86)
- Stochastic Kernel Regularisation Improves Generalisation in Deep Kernel Machines (NeurIPS 2024 · 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.