Research map / Architectures and efficiency / CNN design
Coreset-Based Neural Network Compression
Abhimanyu Dubey, Moitreya Chatterjee, Narendra Ahuja
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 activation-based pruning (AP) to Coreset-S (ablation) | Coreset-S · AlexNet ImageNet | compression ratio: 10.5 -> 19.1 | Improved |
| add activation-based pruning (AP) to Coreset-S (ablation) | Coreset-S · AlexNet ImageNet | top-1 accuracy (%): 56.78 -> 56.38 | Hurt |
| add activation-based pruning (AP) to Coreset-K (ablation) | Coreset-K · AlexNet ImageNet | compression ratio: 6.7 -> 15.2 | Improved |
| add activation-based pruning (AP) to Coreset-A (ablation) | Coreset-A · AlexNet ImageNet | compression ratio: 12.3 -> 16.5 | Improved |
| add activation-based pruning (AP) to Coreset-S (ablation) | Coreset-S · VGGNet-16 ImageNet | compression ratio: 13.9 -> 17 | Improved |
| add activation-based pruning (AP) to Coreset-S (ablation) | Coreset-S · SqueezeNet ImageNet | compression ratio: 1.9 -> 2.1 | Improved |
| couple coreset compression with Deep Compression | AP+Coreset-S · VGGNet-16 ImageNet | compression ratio: 17 -> 233 | Improved |
| couple coreset compression with Deep Compression | AP+Coreset-A · AlexNet ImageNet | compression ratio: 16.5 -> 55 | Improved |
| use activation-based pruning instead of weight-based pruning (ablation) | AP · AlexNet ImageNet | accuracy vs compression | Improved |
| use AP+Coreset-S instead of SVD or pruning when fine-tuning on CUB-2011 | VGG-16 compressed · CUB-2011 | top-1 accuracy (%): 57.45 -> 70.66 | Improved |
| use AP+Coreset-S instead of pruning when fine-tuning on Stanford-Dogs | VGG-16 compressed · Stanford-Dogs | top-1 accuracy (%): 43.28 -> 55.91 | Improved |
| replace retraining-based Dynamic Net Surgery with coresets | compression pipeline · AlexNet ImageNet | training epochs (time units): 140 -> 28 | Improved |
Closest papers by meaning
- Compressing Neural Networks: Towards Determining the Optimal Layer-wise Decomposition (NeurIPS 2021 · similarity 0.91)
- ThiNet: A Filter Level Pruning Method for Deep Neural Network Compression (ICCV 2017 · similarity 0.91)
- Data-Dependent Coresets for Compressing Neural Networks with Applications to Generalization Bounds (ICLR 2019 · similarity 0.90)
- Compressing Deep Convolutional Neural Networks by Stacking Low-dimensional Binary Convolution Filters (AAAI 2021 · similarity 0.90)
- Auto-Compressing Networks (NeurIPS 2025 · similarity 0.88)
- Distribution-Aware Tensor Decomposition for Compression of Convolutional Neural Networks (NeurIPS 2025 · similarity 0.88)
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.
