Research map / Architectures and efficiency / CNN design
Existence, Stability and Scalability of Orthogonal Convolutional Neural 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 |
|---|---|---|---|
| Adding the L_orth orthogonality regularization (lambda=0.1 or 0.01) instead of unconstrained convolutions | VGG-like network on Cifar10 · Cifar10 | clean accuracy and robustness tradeoff | Mixed |
| Adding L_orth regularization on non-square layers (M != CS^2) | synthetic single conv layer · synthetic 64x64 input, all M != CS^2 configurations | singular values sigma_min, sigma_max of K | Helped |
| Using L_orth regularization on square layers (M = CS^2) (ablation) | synthetic single conv layer · synthetic 64x64 input, M=CS^2 configurations | fraction of runs reaching sigma_min~1~sigma_max | Hurt |
| Using circular boundary conditions and L_orth regularization with input size N growing (successful optimization, L_orth<=0.001) | synthetic single conv layer · synthetic, N in {5..1024} | singular values sigma_min and sigma_max | Helped |
| Increasing L_orth regularization strength lambda from 0.01 to 10 on Imagenette | VGG-like network with L_orth · Imagenette | clean accuracy and empirical robustness | Mixed |
| Relaxing orthogonality with lower lambda (less strict L_orth) on Cifar10 | VGG-like network · Cifar10 | clean accuracy | Helped |
Closest papers by meaning
- Orthogonal Convolutional Neural Networks (CVPR 2020 · similarity 0.89)
- An Adaptive Orthogonal Convolution Scheme for Efficient and Flexible CNN Architectures (ICML 2025 · similarity 0.89)
- Scaling-up Diverse Orthogonal Convolutional Networks by a Paraunitary Framework (ICML 2022 · similarity 0.87)
- Constructing Orthogonal Convolutions in an Explicit Manner (ICLR 2022 · similarity 0.87)
- Skew Orthogonal Convolutions (ICML 2021 · similarity 0.85)
- Revisiting Residual Connections: Orthogonal Updates for Stable and Efficient Deep Networks (NeurIPS 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.