Research map / Architectures and efficiency / Pruning and efficient tuning
On the Predictability of Pruning Across Scales
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 |
|---|---|---|---|
| Model pruning error with a five-parameter scaling law using density, depth, width, and unpruned error | IMP-pruned CIFAR-10 ResNets · CIFAR-10 | mean and standard deviation of relative deviation: μ<2%; σ<6% | Helped |
| Use the joint scaling law to predict across width, density, and dataset size | IMP-pruned ImageNet ResNet-50 · ImageNet | mean and standard deviation of relative deviation: μ<2%; σ<6% | Helped |
| Fit the scaling law on smaller widths and depths, then extrapolate to larger networks | IMP-pruned CIFAR-10 ResNets · CIFAR-10 | mean and standard deviation of relative deviation: μ<7%, σ<6%; μ<1%, σ<6% | Mixed |
| Fit from only 40 randomly sampled network points | Joint scaling law on CIFAR-10 ResNets · CIFAR-10 | fit stability: one percentage point | Helped |
| Fit from 15 sampled configurations, including all densities for each | Joint scaling law on CIFAR-10 ResNets · CIFAR-10 | fit stability: below 1% | Helped |
| Use a pruning-specific rational functional form rather than the adapted non-pruning scaling form | Pruning error modeling · CIFAR-10 ResNets | fit to error versus density | Helped |
| Treat the low-error plateau as flat, without modeling error dips below unpruned error | Joint pruning scaling law · IMP-pruned networks | relative bias: ~1% | Hurt |
Closest papers by meaning
- Pruning from Scratch (AAAI 2020 · similarity 0.91)
- Neural Network Pruning Denoises the Features and Makes Local Connectivity Emerge in Visual Tasks (ICML 2022 · similarity 0.91)
- Pruning Neural Networks at Initialization: Why Are We Missing the Mark? (ICLR 2021 · similarity 0.91)
- How Sparse Can We Prune A Deep Network: A Fundamental Limit Perspective (NeurIPS 2024 · similarity 0.89)
- Distilling the Knowledge in Data Pruning (ICML 2025 · similarity 0.88)
- Sparsest Models Elude Pruning: An Exposé of Pruning’s Current Capabilities (ICML 2024 · 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.