Research map / Architectures and efficiency / Pruning and efficient tuning
RED++ : Data-Free Pruning of Deep Neural Networks via Input Splitting and Output Merging
Edouard Yvinec, Arnaud Dapogny, Matthieu Cord, Kévin Bailly
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 input splitting step (hashing + merge alpha*) (ablation) | RED++ · Cifar10 ResNet 56 | pruning ratio (% params removed): 77.68 -> 88.81 | Improved |
| add input splitting step (hashing + merge alpha*) (ablation) | RED++ · Cifar10 ResNet 20 | pruning ratio (% params removed): 41.03 -> 67.48 | Improved |
| add input splitting step (hashing + merge alpha*) (ablation) | RED++ · ImageNet ResNet 50 | pruning ratio (% params removed): 0.29 -> 44.25 | Improved |
| add weight hashing to merge alpha* and split (ablation) | RED++ · Cifar10 ResNet 56 | pruning ratio (% params removed): 61.19 -> 88.81 | Improved |
| add weight hashing to merge alpha* and split (ablation) | RED++ · ImageNet ResNet 50 | pruning ratio (% params removed): 0.09 -> 44.25 | Improved |
| use relaxed merging alpha* instead of alpha=0 (with hashing and split) (ablation) | RED++ · Cifar10 ResNet 56 | pruning ratio (% params removed): 85.89 -> 88.81 | Improved |
| use relaxed merging alpha* instead of alpha=0 (with hashing and split) (ablation) | RED++ · Cifar10 ResNet 20 | pruning ratio (% params removed): 65.26 -> 67.48 | Improved |
| merge alpha=0 without hashing (ablation) | RED++ · Cifar10 and ImageNet networks | pruning ratio (% params removed): 0.00 | No clear effect |
| use adaptive hashing instead of uniform int8 hashing (ablation) | RED++ · Cifar10 ResNet 56 | % accuracy drop: 0.45 -> -0.03 | Improved |
| use adaptive hashing instead of uniform int8 hashing (ablation) | RED++ · ImageNet MobileNet V2 (0.35) | % accuracy drop: 3.96 -> 0.25 | Improved |
| use adaptive hashing instead of uniform on EfficientNet B1 (ablation) | RED++ · ImageNet EfficientNet B1 | % accuracy drop: 14.32 -> 0.36 | Improved |
| use block alpha strategy instead of constant alpha (ablation) | RED++ · Cifar10 ResNet 56 | pruning factor (%): 72.40 -> 77.68 | Improved |
Closest papers by meaning
- RED : Looking for Redundancies for Data-FreeStructured Compression of Deep Neural Networks (NeurIPS 2021 · similarity 0.95)
- 1xN Pattern for Pruning Convolutional Neural Networks (TPAMI 2022 · similarity 0.89)
- Rethinking the Value of Network Pruning (ICLR 2019 · similarity 0.89)
- Convolutional Neural Network Pruning With Structural Redundancy Reduction (CVPR 2021 · similarity 0.88)
- Accelerate CNNs from Three Dimensions: A Comprehensive Pruning Framework (ICML 2021 · similarity 0.88)
- Structured Pruning for Deep Convolutional Neural Networks: A Survey (TPAMI 2023 · 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.
