Research map / Architectures and efficiency / CNN design
CNNpack: Packing Convolutional Neural Networks in the Frequency Domain
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 cluster centers (K=16) shared across filters in the frequency domain (ablation) | AlexNet compression · ImageNet ILSVRC-2012 | compression ratio: from close to 35× (K=0) to 39× | Helped |
| add cluster centers (K=16) shared across filters in the frequency domain (ablation) | AlexNet compression · ImageNet ILSVRC-2012 | speed-up ratio: from close to 22× (K=0) to 25× | Hurt |
| lower DCT filter size d below max filter size (ablation) | CNNpack on MNIST LeNet · MNIST | accuracy | Hurt |
| apply bound b on quantized values (b=0.05 instead of +inf) (ablation) | CNNpack on MNIST LeNet · MNIST | accuracy | Hurt |
| increase lambda (sparsity penalty) from 0.025 to 0.055 (ablation) | CNNpack on MNIST LeNet · MNIST | compression ratio and speed-up | Helped |
| CNNpack compression with lambda=0.04 and K=16 vs pruning-based and P+QH baselines on AlexNet | AlexNet · ImageNet ILSVRC-2012 | compression ratio: CNNpack 39× vs P+QH 35× and XNOR 64× | Mixed |
| use DCT-domain compression with clustering on VGG-16 | VGG-16 · ImageNet ILSVRC-2012 | compression ratio: CNNpack 46× vs P+QH 49× | Mixed |
| compress ResNet-50 in DCT domain with K=0 (no clustering), discarding ~84% of frequency weights | ResNet-50 · ImageNet ILSVRC-2012 | memory compression | Helped |
Closest papers by meaning
- Frequency-Domain Dynamic Pruning for Convolutional Neural Networks (NeurIPS 2018 · similarity 0.89)
- Multiscale Low-Frequency Memory Network for Improved Feature Extraction in Convolutional Neural Networks (AAAI 2024 · similarity 0.89)
- Compressing Deep Convolutional Neural Networks by Stacking Low-dimensional Binary Convolution Filters (AAAI 2021 · similarity 0.89)
- Frequency Domain Compact 3D Convolutional Neural Networks (CVPR 2020 · similarity 0.88)
- Frequency-Biased Synergistic Design for Image Compression and Compensation (CVPR 2025 · similarity 0.87)
- Distribution-Aware Tensor Decomposition for Compression of Convolutional Neural Networks (NeurIPS 2025 · similarity 0.86)
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.