Research map / Architectures and efficiency / CNN design
Deeply Shared Filter Bases for Parameter-Efficient 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 |
|---|---|---|---|
| Add orthogonality regularization to the shared filter basis (ablation) | ResNet50-Shared · ImageNet val | top-1 error (%): 24.19 -> 23.64 | Helped |
| Keep coefficients unshared rather than sharing them recursively (ablation) | ResNet50-Shared · ImageNet val | top-1 error (%): 24.35 -> 23.64 | Helped |
| Use two shared bases per residual block group instead of one (ablation) | ResNet56-S16U1 · CIFAR-10 test | error (%): 7.46 -> 6.33 | Helped |
| Add orthogonality regularization (ablation) | ResNet34-S32U1 · CIFAR-100 test | error (%): 22.92 -> 21.84 | Helped |
| Add orthogonality regularization (ablation) | ResNet56-S16U1 · CIFAR-10 test | error (%): 7.70 -> 7.46 | Helped |
| Share filter bases recursively while retaining non-shared coefficients | ResNet34 · CIFAR-10 test | error (%): 6.97 -> 7.46 | Hurt |
| Share filter bases recursively with orthogonality regularization | ResNet50 · ImageNet val | top-1 error (%): 23.85 -> 23.64 | Helped |
| Share filter bases recursively with orthogonality regularization | ResNet101 · ImageNet val | top-1 error (%): 22.63 -> 22.31 | Helped |
| Share filter bases recursively with orthogonality regularization | MobileNetV2 · ImageNet val | top-1 error (%): 28.0 -> 27.61 | Helped |
| Share filter bases recursively | DenseNet121 · CIFAR-100 test | error (%): 21.95 -> 22.15 | Hurt |
| Share filter bases recursively with four non-shared basis elements | ResNeXt50 · CIFAR-100 test | error (%): 20.71 -> 20.09 | Helped |
Closest papers by meaning
- Scaling-up Diverse Orthogonal Convolutional Networks by a Paraunitary Framework (ICML 2022 · similarity 0.88)
- RePr: Improved Training of Convolutional Filters (CVPR 2019 · similarity 0.88)
- Learning Implicitly Recurrent CNNs Through Parameter Sharing (ICLR 2019 · similarity 0.88)
- CNN Filter DB: An Empirical Investigation of Trained Convolutional Filters (CVPR 2022 · similarity 0.87)
- Convexified Convolutional Neural Networks (ICML 2017 · similarity 0.87)
- PICNN: A Pathway towards Interpretable Convolutional Neural Networks (AAAI 2024 · 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.