Research map / Architectures and efficiency / Pruning and efficient tuning
Towards Data-Agnostic Pruning At Initialization: What Makes a Good Sparse Mask?
Hoang Pham, The-Anh Ta, Shiwei Liu, Lichuan Xiang, Dung D. Le, Hongkai Wen, Long Tran-Thanh
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 |
|---|---|---|---|
| Use NPB node-path balancing instead of SNIP | SNIP · VGG-19 CIFAR-100, 99% sparsity | accuracy (%): 61.77 -> 64.82 | Improved |
| Use NPB instead of SynFlow | SynFlow · ResNet-20 CIFAR-10, 99% sparsity | accuracy (%): 66.05 -> 70.70 | Improved |
| Use NPB instead of Iter-SNIP | Iter-SNIP · ResNet-18 Tiny-ImageNet, 99% sparsity | accuracy (%): 36.42 -> 41.05 | Improved |
| Use NPB instead of PHEW | PHEW · ResNet-18 Tiny-ImageNet, 90% sparsity | accuracy (%): 55.93 -> 56.82 | Improved |
| Use NPB instead of PHEW | PHEW · ResNet-20 CIFAR-10, 99% sparsity | accuracy (%): 70.44 -> 70.70 | No clear effect |
| Use NPB instead of PHEW | PHEW · ResNet-20 CIFAR-10, 96.84% sparsity | accuracy (%): 81.05 -> 80.55 | No clear effect |
| Use NPB instead of PHEW for pruning time | PHEW · ResNet-18 Tiny-ImageNet, 90% sparsity | pruning time (s): 1342.03 -> 375.65 | Improved |
| Use NPB instead of SynFlow for pruning time | SynFlow · ResNet-18 Tiny-ImageNet, 90% sparsity | pruning time (s): 96.18 -> 375.65 | Hurt |
| Use NPB instead of PHEW for inference cost | PHEW · ResNet-18 Tiny-ImageNet, 99% sparsity | FLOPs (1e8): 1.50 -> 0.59 | Improved |
| Use NPB instead of SNIP | SNIP · Three settings, various sparsities | accuracy: up to 14% | Improved |
| Use NPB instead of SynFlow | SynFlow · Three settings, various sparsities | accuracy: up to 7% | Improved |
| Random pruning at sparsity above 90% | Random pruning · Three settings | accuracy | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SNIP — SNIP: SINGLE-SHOT NETWORK PRUNING BASED ON CONNECTION SENSITIVITY (ICLR 2019) · better on 6 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- REVISITING PRUNING AT INITIALIZATION THROUGH THE LENS OF RAMANUJAN GRAPH (ICLR 2023 · similarity 0.93)
- DPaI: Differentiable Pruning at Initialization with Node-Path Balance Principle (ICLR 2025 · similarity 0.91)
- Pruning Neural Networks at Initialization: Why Are We Missing the Mark? (ICLR 2021 · similarity 0.91)
- Pruning neural networks without any data by iteratively conserving synaptic flow (NeurIPS 2020 · similarity 0.90)
- Connectivity Matters: Neural Network Pruning Through the Lens of Effective Sparsity (JMLR 2023 · similarity 0.90)
- No Free Prune: Information-Theoretic Barriers to Pruning at Initialization (ICML 2024 · similarity 0.89)
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.
