Research map / Architectures and efficiency / CNN design
Interleaved Structured Sparse 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 |
|---|---|---|---|
| Replace Xception blocks with IGC-V2* blocks | 20-layer network · CIFAR-100 | accuracy: 74.46 ± 0.24 -> 75.35 ± 0.56 | Helped |
| Replace Xception blocks with IGC-V2* blocks | 20-layer network · Tiny ImageNet | accuracy: 57.34 ± 0.34 -> 59.40 ± 0.22 | Helped |
| Replace IGC-V1 blocks with IGC-V2* blocks | 20-layer network · CIFAR-100 | accuracy: 75.10 ± 0.36 -> 75.35 ± 0.56 | Helped |
| Replace IGC-V1 blocks with IGC-V2* blocks | 20-layer network · Tiny ImageNet | accuracy: 58.37 ± 0.60 -> 59.40 ± 0.22 | Helped |
| Use IGC-V2 instead of Xception at depth 26 | network · CIFAR-100 | accuracy: 72.48 ± 0.18 -> 72.94 ± 0.26 | Helped |
| Use IGC-V2 instead of Xception at depth 26 | network · Tiny ImageNet | accuracy: 55.39 ± 0.35 -> 56.32 ± 0.38 | Helped |
| Use IGC-V2 instead of IGC-V1 at depth 26 | network · CIFAR-100 | accuracy: 71.82 ± 0.40 -> 73.49 ± 0.34 | Helped |
| Use IGC-V2 instead of IGC-V1 at depth 26 | network · Tiny ImageNet | accuracy: 56.64 ± 0.15 -> 57.12 ± 0.09 | Helped |
| Use a channel-wise spatial convolution followed by two balanced group 1×1 convolutions instead of the IGC-V1-like design | 8-layer IGC-V2 block · CIFAR-100 | accuracy: 61.35 ± 0.41 -> 62.32 ± 0.57 | Helped |
| Use a channel-wise spatial convolution followed by two balanced group 1×1 convolutions instead of the IGC-V1-like design | 8-layer IGC-V2 block · CIFAR-10 | accuracy: 87.41 ± 0.35 -> 88.01 ± 0.32 | Helped |
| Increase the number of group convolutions L while maintaining parameter count (ablation) | IGC-V2 block · CIFAR-100 | accuracy | Mixed |
| Use a wider network instead of a deeper network (ablation) | IGC-V2* network · CIFAR-100 | accuracy | Helped |
Closest papers by meaning
- A General Efficient Hyperparameter-Free Algorithm for Convolutional Sparse Learning (AAAI 2017 · similarity 0.88)
- Structured Convolutions for Efficient Neural Network Design (NeurIPS 2020 · similarity 0.87)
- Exploiting Redundancy: Separable Group Convolutional Networks on Lie Groups (ICML 2022 · similarity 0.86)
- Revisiting Sparse Convolutional Model for Visual Recognition (NeurIPS 2022 · similarity 0.86)
- Revisit Kernel Pruning with Lottery Regulated Grouped Convolutions (ICLR 2022 · similarity 0.86)
- SequentialAttention++ for Block Sparsification: Differentiable Pruning Meets Combinatorial Optimization (NeurIPS 2024 · similarity 0.84)
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.