Research map / Architectures and efficiency / Pruning and efficient tuning
Pruning Randomly Initialized Neural Networks with Iterative Randomization
Daiki Chijiwa, Shin'ya Yamaguchi, Yasutoshi Ida, Kenji Umakoshi, Tomohiro INOUE
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 iterative partial randomization of pruned weights to edge-popup | Conv6 · CIFAR-10 test, ρ=0.25, SC | accuracy: 70.35 ± 1.16 -> 75.16 ± 0.23 | Improved |
| Add iterative partial randomization of pruned weights to edge-popup | ResNet18 · CIFAR-10 test, ρ=0.25, SC | accuracy: 84.23 ± 0.33 -> 86.09 ± 0.24 | Improved |
| Add iterative randomization during pruning | ResNet34 · ImageNet validation, 70% sparsity | Top-1 accuracy: 62.14 -> 65.86 | Improved |
| Add iterative randomization during pruning | ResNet18 · CIFAR-10 test, p=0.99, SC | accuracy: 75.21 ± 2.71 -> 76.31 ± 2.56 | No clear effect |
| Use IMP with 3 retraining instead of pruning-only IteRand | VGG11 · CIFAR-10, p=0.5 | accuracy: 88.46 ± 0.22 -> 91.47 ± 0.15 | Improved |
| Use partial randomization with r=0.1 rather than the other tested rates (ablation) | CNNs · CIFAR-10 validation | accuracy | Improved |
| Vary K_per; avoid very frequent randomization and periods so long that randomization nearly disappears (ablation) | IteRand · CIFAR-10 validation, Conv6/ResNet18/ResNet34 | accuracy | Mixed |
| Increase r and adjust K_per to preserve the approximate randomization rate (ablation) | IteRand · CIFAR-10 test, ResNet18 | accuracy | No clear effect |
| Increase width factor ρ (ablation) | IteRand and edge-popup · CIFAR-10, Conv6/ResNet18/ResNet34 | accuracy | Improved |
| Use IteRand instead of edge-popup at matched width | ResNet18, KU · CIFAR-10 test, ρ=0.5 | accuracy: 88.45± 0.53 -> 89.96± 0.06 | Improved |
| Add iterative randomization during pruning | LSTM · IMDB test | accuracy: 88.16 ± 0.12 % -> 88.44 ± 0.28 % | Improved |
Closest papers by meaning
- Pruning Neural Networks at Initialization: Why Are We Missing the Mark? (ICLR 2021 · similarity 0.89)
- Validating the Lottery Ticket Hypothesis with Inertial Manifold Theory (NeurIPS 2021 · similarity 0.89)
- The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks (ICLR 2019 · similarity 0.88)
- Pruning at Initialization – A Sketching Perspective (TPAMI 2025 · similarity 0.88)
- Polynomially Over-Parameterized Convolutional Neural Networks Contain Structured Strong Winning Lottery Tickets (NeurIPS 2023 · similarity 0.88)
- No Free Prune: Information-Theoretic Barriers to Pruning at Initialization (ICML 2024 · similarity 0.87)
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.
