Research map / Architectures and efficiency / CNN design
Revisiting Spatial Invariance with Low-Rank Local Connectivity
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 |
|---|---|---|---|
| Replace the inserted convolution layer with LRLC | ResNet-50 · ImageNet validation, insert layer | top-1 accuracy: 77.22±0.03 -> 77.47±0.03 | Helped |
| Replace 3×3 convolutions with LRLC | ResNet-50 · ImageNet validation, replace 3×3 convolutions | top-1 accuracy: 76.93±0.07 -> 77.08±0.02 | Helped |
| Replace 3×3 convolutions with input-dependent LRLC | ResNet-50 · ImageNet validation, replace 3×3 convolutions | top-1 accuracy: 76.93±0.07 -> 77.80±0.02 | Helped |
| Use input-dependent rather than fixed combining weights | LRLC · CIFAR-10 | top-1 accuracy | Helped |
| Use input-dependent rather than fixed combining weights | LRLC · MNIST | top-1 accuracy | No clear effect |
| Use input-dependent rather than fixed combining weights | LRLC · CelebA | top-1 accuracy | Hurt |
| Relax spatial invariance in later rather than earlier network layers (ablation) | LRLC network · MNIST, CIFAR-10, and CelebA | top-1 accuracy | Helped |
| Relax spatial invariance in later layers rather than at every layer (ablation) | LRLC network · MNIST, CIFAR-10, and CelebA | top-1 accuracy | Helped |
| Add spatially varying bias to convolution (rank-1 LRLC) (ablation) | Convolutional network · MNIST, CIFAR-10, and CelebA | top-1 accuracy | Helped |
| Replace factorized combining weights and biases with unfactorized weights and biases (ablation) | LRLC · Classification datasets | top-1 accuracy | Hurt |
Closest papers by meaning
- Tilt your Head: Activating the Hidden Spatial-Invariance of Classifiers (ICML 2024 · similarity 0.88)
- Role of Locality and Weight Sharing in Image-Based Tasks: A Sample Complexity Separation between CNNs, LCNs, and FCNs (ICLR 2024 · similarity 0.87)
- From Bricks to Bridges: Product of Invariances to Enhance Latent Space Communication (ICLR 2024 · similarity 0.85)
- The Implicit Bias of Heterogeneity towards Invariance: A Study of Multi-Environment Matrix Sensing (NeurIPS 2024 · similarity 0.85)
- A Probabilistic Approach to Learning the Degree of Equivariance in Steerable CNNs (ICML 2024 · similarity 0.85)
- Locality in Image Diffusion Models Emerges from Data Statistics (NeurIPS 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.