Research map / Architectures and efficiency / CNN design
Structure-Aware 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 |
|---|---|---|---|
| learn local structure representations instead of fixed uniformly sampled ones (ablation) | SACNNs · STL-10 | accuracy: 0.7175 -> 0.7358 | Helped |
| learn local structure representations instead of fixed uniformly sampled ones (ablation) | SACNNs · Cifar-10 | accuracy: 0.9091 -> 0.9167 | Helped |
| learn local structure representations instead of fixed uniformly sampled ones (ablation) | SACNNs · Mnist | accuracy: 0.9957 -> 0.9961 | No clear effect |
| learn local structure representations instead of fixed uniformly sampled ones (ablation) | SACNNs · 20News | accuracy: 0.7362 -> 0.7436 | Helped |
| learn local structure representations instead of fixed uniformly sampled ones (ablation) | SACNNs · TF-198 | RMSE: 58.82 -> 53.72 | Helped |
| learn local structure representations instead of fixed uniformly sampled ones (ablation) | SACNNs · ImageDog (DAC clustering) | accuracy: 0.2801 -> 0.2913 | Helped |
| replace classical 3x3 convolution with structure-aware convolution (SACNNs) vs ClaCNNs | CNN · STL-10 | accuracy: 0.6635 -> 0.7358 | Helped |
| replace classical convolution with structure-aware convolution (SACNNs) vs ClaCNNs | CNN · Mnist | accuracy: 0.9953 -> 0.9961 | No clear effect |
| increase Chebyshev polynomial order t (5 to 80+) (ablation) | SACNNs 11x11 filters · STL-10 | validation error | Helped |
| use Legendre instead of Chebyshev basis (ablation) | SACNNs · Cifar-10 | training curve | No clear effect |
| replace classical convolution in ResNets with structure-aware convolution | ResNet · Cifar-10 | testing error % | Helped |
| pre-train on Reuters before fine-tuning on 20News (ablation) | SACNNs · 20News (1k-3k labels) | validation error | Helped |
Closest papers by meaning
- Structured Convolutions for Efficient Neural Network Design (NeurIPS 2020 · similarity 0.89)
- Linear CNNs Discover the Statistical Structure of the Dataset Using Only the Most Dominant Frequencies (ICML 2023 · similarity 0.89)
- Constructing Deep Neural Networks by Bayesian Network Structure Learning (NeurIPS 2018 · similarity 0.87)
- Unveiling the Unseen: Identifiable Clusters in Trained Depthwise Convolutional Kernels (ICLR 2024 · similarity 0.85)
- PICNN: A Pathway towards Interpretable Convolutional Neural Networks (AAAI 2024 · similarity 0.85)
- AsCAN: Asymmetric Convolution-Attention Networks for Efficient Recognition and Generation (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.