Research map / Architectures and efficiency / CNN design
CircConv: A Structured Convolution with Low Complexity
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 AlexNet convolutional layers with CircConv using configuration 1-2-2-2-2 | AlexNet · ImageNet ILSVRC-2012 | test error (%): 42.9 -> 42.75 | Helped |
| Replace AlexNet convolutional layers with CircConv using configuration 1-2-2-2-2 | AlexNet · ImageNet ILSVRC-2012 | FLOPs (%): 100 -> 31.3 | Helped |
| Use configuration 1-2-2-4-2 instead of 1-2-2-2-2 (ablation) | CircConv AlexNet · ImageNet ILSVRC-2012 | parameters (%): 50.36 -> 40.01 | Helped |
| Use configuration 1-2-2-4-2 instead of 1-2-2-2-2 (ablation) | CircConv AlexNet · ImageNet ILSVRC-2012 | test error (%): 42.75 -> 42.99 | Hurt |
| Use configuration 1-2-4-2-2 instead of 1-2-2-2-2 (ablation) | CircConv AlexNet · ImageNet ILSVRC-2012 | test error (%): 42.75 -> 43.13 | Hurt |
| Use configuration 1-2-4-2-2 instead of 1-2-2-2-2 (ablation) | CircConv AlexNet · ImageNet ILSVRC-2012 | parameters (%): 50.36 -> 45.19 | Helped |
| Train the circulant model from scratch rather than convert and retrain a pretrained model (ablation) | CircConv CNNs · CIFAR-10 and ImageNet ILSVRC-2012 | test accuracy | No clear effect |
| Apply CircConv compression to ResNet-32 model 1 | ResNet-32 · CIFAR-10 | test error | Helped |
| Increase ResNet-32 compression to configurations 2, 3, and 4 (ablation) | ResNet-32 · CIFAR-10 | test error | No clear effect |
| Use more aggressive ResNet-32 compression configurations 5, 6, and 7 (ablation) | ResNet-32 · CIFAR-10 | test error | Hurt |
| Apply CircConv to Wide ResNet convolutional layers with matching input and output channels | Wide ResNet · CIFAR-10 | test error | Hurt |
| Apply CircConv to Wide ResNet convolutional layers with matching input and output channels | Wide ResNet · CIFAR-10 | FLOPs: around 36% | Helped |
Closest papers by meaning
- Circulant Binary Convolutional Networks: Enhancing the Performance of 1-Bit DCNNs With Circulant Back Propagation (CVPR 2019 · similarity 0.88)
- Structured Convolutions for Efficient Neural Network Design (NeurIPS 2020 · similarity 0.83)
- CondConv: Conditionally Parameterized Convolutions for Efficient Inference (NeurIPS 2019 · similarity 0.82)
- ClcNet: Improving the Efficiency of Convolutional Neural Network Using Channel Local Convolutions (CVPR 2018 · similarity 0.82)
- The Unreasonable Effectiveness of Fully-Connected Layers for Low-Data Regimes (NeurIPS 2022 · similarity 0.82)
- Deep Tensor Convolution on Multicores (ICML 2017 · similarity 0.82)
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.