Research map / Architectures and efficiency / Pruning and efficient tuning
One-Shot Pruning of Recurrent Neural Networks by Jacobian Spectrum Evaluation
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 Jacobian-spectrum pruning instead of random pruning | GRU · Sequential MNIST, 400 units, 95% sparsity | validation error (%): 1.50±0.08 -> 1.46±0.05 | Helped |
| use Jacobian-spectrum pruning instead of random pruning | GRU · Sequential MNIST, 100 units, 95% sparsity | validation error (%): 2.78±0.25 -> 3.09±0.31 | Hurt |
| use Jacobian-spectrum pruning instead of random pruning | GRU · Sequential MNIST, 1600 units, 95% sparsity | validation error (%): 1.15±0.12 -> 1.01±0.05 | Helped |
| use Jacobian-spectrum pruning instead of random pruning | LSTM · WikiText-2 | training perplexity: 22.66 -> 20.54 | Helped |
| use Jacobian-spectrum pruning instead of random pruning | LSTM · WikiText-103 | training perplexity: 49.65 -> 46.65 | Helped |
| use Jacobian-spectrum pruning instead of random pruning | LSTM · Truncated One Billion Words | training perplexity: 59.17 -> 53.26 | Helped |
| add normalization across the pruning objective (ablation) | SNIP and Foresight pruning criteria on recurrent networks · Sequential MNIST, 95% sparse GRU, 400 units | validation error (%): 88.8±0.1 -> 1.52±0.11 | Helped |
| use Glorot initialization (ablation) | Jacobian-spectrum pruning on GRU · Sequential MNIST, 400 units | validation error (%): 1.36 -> 1.219 | Helped |
| use N(0,1) initialization (ablation) | Jacobian-spectrum pruning on GRU · Sequential MNIST, 400 units | validation error (%): 1.38 -> 3.30 | Hurt |
| use uniform(0,0.1) initialization (ablation) | Jacobian-spectrum pruning on GRU · Sequential MNIST, 400 units | validation error (%): 1.32 -> 1.73 | Hurt |
| prune 95% of parameters rather than retain all parameters (ablation) | 800-unit GRU · Penn Treebank | validation perplexity: 156.16 -> 208.14 | Hurt |
Closest papers by meaning
- Targeted Low-rank Refinement: Enhancing Sparse Language Models with Precision (ICML 2025 · similarity 0.90)
- Sparse Spiking Neural Network: Exploiting Heterogeneity in Timescales for Pruning Recurrent SNN (ICLR 2024 · similarity 0.88)
- SAFE: Finding Sparse and Flat Minima to Improve Pruning (ICML 2025 · similarity 0.88)
- Sparsest Models Elude Pruning: An Exposé of Pruning’s Current Capabilities (ICML 2024 · similarity 0.88)
- The Journey Matters: Average Parameter Count over Pre-training Unifies Sparse and Dense Scaling Laws (ICLR 2025 · similarity 0.88)
- Olica: Efficient Structured Pruning of Large Language Models without Retraining (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.