Research map / Architectures and efficiency / CNN design
Convolutional Neural Network Compression through Generalized Kronecker Product Decomposition
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 ResNet32 tensor decomposition with GKPD at approximately 5× compression | ResNet32 · CIFAR-10 | top-1 accuracy (%): 90.6 -> 91.52 | Helped |
| Replace Tucker decomposition with GKPD at approximately 5× compression | ResNet32 · CIFAR-10 | top-1 accuracy (%): 87.7 -> 91.52 | Helped |
| Replace Tensor-Train decomposition with GKPD at similar compression | ResNet32 · CIFAR-10 | top-1 accuracy (%): 88.3 -> 91.52 | Helped |
| Use GKPD instead of the strongest listed pruning or filter-sharing baseline at roughly 2× compression | ResNet50 · ImageNet | top-1 accuracy (%): 73.11 -> 73.95 | Helped |
| Compress ResNet18 with GKPD to approximately 5× fewer parameters | ResNet18 · CIFAR-10 | top-1 accuracy (%): 95.05 -> 94.97 | Hurt |
| Compress SEResNet50 with GKPD to a similar parameter count to MobileNetV2 | SEResNet50 · CIFAR-10 | top-1 accuracy (%): 95.15 -> 94.45 | Hurt |
| Increase the GKPD rank from 4 to 8 products per layer (ablation) | KroneckerResNet18 · CIFAR-10 | top-1 accuracy (%): 92.96 -> 93.74 | Helped |
| Increase the GKPD rank from 8 to 16 products per layer (ablation) | KroneckerResNet18 · CIFAR-10 | top-1 accuracy (%): 93.74 -> 94.30 | Helped |
| Increase the GKPD rank from 16 to 32 products per layer (ablation) | KroneckerResNet18 · CIFAR-10 | top-1 accuracy (%): 94.30 -> 94.58 | Helped |
Closest papers by meaning
- Exploring Unexplored Tensor Network Decompositions for Convolutional Neural Networks (NeurIPS 2019 · similarity 0.88)
- Compressing Neural Networks: Towards Determining the Optimal Layer-wise Decomposition (NeurIPS 2021 · similarity 0.88)
- Distribution-Aware Tensor Decomposition for Compression of Convolutional Neural Networks (NeurIPS 2025 · similarity 0.87)
- Towards Compact CNNs via Collaborative Compression (CVPR 2021 · similarity 0.87)
- Attention-based Iterative Decomposition for Tensor Product Representation (ICLR 2024 · similarity 0.85)
- Convolutions and More as Einsum: A Tensor Network Perspective with Advances for Second-Order Methods (NeurIPS 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.