Research map / Architectures and efficiency / Pruning and efficient tuning
Dynamic Model Pruning with Feedback
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 |
|---|---|---|---|
| use dynamic pruning with feedback instead of Incremental pruning | ResNet-50 · ImageNet, 80% target pruning | top-1 accuracy: 74.25 -> 75.48 | Helped |
| use dynamic pruning with feedback instead of Incremental pruning | ResNet-50 · ImageNet, 90% target pruning | top-1 accuracy: 73.36 -> 74.55 | Helped |
| add 60 epochs of fine-tuning (ablation) | Dynamic pruning with feedback on ResNet-20 · CIFAR-10, 95% target pruning | top-1 test accuracy: 88.01 \pm 0.30 -> 90.34 \pm 0.31 | Helped |
| add 60 epochs of fine-tuning (ablation) | Dynamic pruning with feedback on ResNet-56 · CIFAR-10, 90% target pruning | top-1 test accuracy: 93.95 \pm 0.11 -> 93.95 \pm 0.17 | No clear effect |
| use layer-wise instead of global-wise reparameterization (ablation) | Dynamic pruning with feedback on ResNet-20 · CIFAR-10, 50% target sparsity | MACs: 22.80M -> 20.99M | Helped |
| increase the reparameterization period from the reported optimum (ablation) | Dynamic pruning with feedback · CIFAR-10; varied sparsity and mini-batch sizes | top-1 test accuracy | Hurt |
| apply feedback updates to the dense weights while computing gradients at the pruned model | Dynamic pruning with feedback · CIFAR-10 and ImageNet pruning experiments | top-1 accuracy | Helped |
| search for a fixed-size subnetwork from a larger WideResNet (ablation) | Dynamic pruning with feedback · CIFAR-10, WideResNet-28 | top-1 accuracy | Helped |
Closest papers by meaning
- Computation and Memory-Efficient Model Compression with Gradient Reweighting (NeurIPS 2025 · similarity 0.91)
- SPDY: Accurate Pruning with Speedup Guarantees (ICML 2022 · similarity 0.90)
- Targeted Low-rank Refinement: Enhancing Sparse Language Models with Precision (ICML 2025 · similarity 0.89)
- Sparsest Models Elude Pruning: An Exposé of Pruning’s Current Capabilities (ICML 2024 · similarity 0.88)
- Sparse Model Soups: A Recipe for Improved Pruning via Model Averaging (ICLR 2024 · similarity 0.88)
- Numerical Pruning for Efficient Autoregressive Models (AAAI 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.