Research map / Architectures and efficiency / Pruning and efficient tuning
Why Random Pruning Is All We Need to Start Sparse
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 balanced rather than uniform initial layerwise sparsity for STR pruning (ablation) | ER-start STR on ResNet18 · CIFAR10, final sparsity 0.997 | test accuracy: 88.92 -> 90.7 | Helped |
| Use pyramidal rather than ERK initial layerwise sparsity for IMP pruning (ablation) | ER-start IMP on ResNet18 · CIFAR10, final sparsity 0.9 | test accuracy: 93.5 -> 93.65 | Helped |
| Start IMP from an ER mask with pyramidal sparsity instead of a dense network | IMP on ResNet18 · CIFAR10, final sparsity 0.9 | test accuracy: 93.38 -> 93.65 | Helped |
| Start edge-popup from an ER mask with sparsity 0.8 rather than a dense network | Edge-popup on ResNet18 · CIFAR10, final sparsity 0.95 | test accuracy: 87.86 -> 88.1 | Helped |
| Rewire ERK edges with RiGL (ablation) | ERK VGG16 · CIFAR10, sparsity 0.99 | test accuracy: 87.8 -> 90.8 | Helped |
| Rewire balanced ER edges with RiGL (ablation) | Balanced ER VGG16 · CIFAR10, sparsity 0.99 | test accuracy: 89.3 -> 91.4 | Helped |
| Rewire pyramidal ER edges with RiGL (ablation) | Pyramidal ER VGG16 · CIFAR10, sparsity 0.999 | test accuracy: 10 -> 9.8 | Hurt |
| Use balanced rather than pyramidal layerwise sparsity (ablation) | Random ER VGG16 · CIFAR10, sparsity 0.999 | test accuracy: 10 -> 68.7 | Helped |
| Increase random ER sparsity to 0.999 rather than 0.99 (ablation) | Pyramidal ER VGG16 · CIFAR10 | test accuracy: 90.4 -> 10 | Hurt |
| Start STR pruning from a random ER mask rather than a dense network | STR on ResNet18 · CIFAR10, final sparsity 0.96 | test accuracy: 94.66 -> 94.11 | Hurt |
Closest papers by meaning
- The Unreasonable Effectiveness of Random Pruning: Return of the Most Naive Baseline for Sparse Training (ICLR 2022 · similarity 0.94)
- No Free Prune: Information-Theoretic Barriers to Pruning at Initialization (ICML 2024 · similarity 0.92)
- Training Your Sparse Neural Network Better with Any Mask (ICML 2022 · similarity 0.92)
- Sparsest Models Elude Pruning: An Exposé of Pruning’s Current Capabilities (ICML 2024 · similarity 0.89)
- How Sparse Can We Prune A Deep Network: A Fundamental Limit Perspective (NeurIPS 2024 · similarity 0.89)
- Sparse Training from Random Initialization: Aligning Lottery Ticket Masks using Weight Symmetry (ICML 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.