Research map / Architectures and efficiency / Pruning and efficient tuning
Layerwise Sparse Coding for Pruned Deep Neural Networks with Extreme Compression Ratio
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 |
|---|---|---|---|
| add block layer (zero-block removal) before SRI coding (ablation) | SRI · ADMM-Lenet average | compression ratio: 33.04x -> 51.03x | Helped |
| add block layer before SRI coding (ablation) | SRI · Inference, 98% sparse matrix | time (ms): 81.65 -> 19.34 | Helped |
| add block layer before SRI coding (ablation) | SRI · Inference, 98% sparse matrix | memory (MB): 2.37 -> 0.10 | Helped |
| add block layer on low-sparsity layer conv1 (0.8) (ablation) | SRI · ADMM-Lenet conv1 | compression ratio: 4.27x -> 4.03x | Hurt |
| replace Bitmask with SRI in coding layer (ablation) | LSC · Random matrices, 99% sparse, 3x3 | compression ratio: 61.83x -> 66.05x | Helped |
| replace Bitmask with SRI in coding layer (ablation) | LSC · Random matrices, 90% sparse, 3x3 | compression ratio: 8.25x -> 8.58x | Helped |
| increase block size from 3x3 to 4x4 (ablation) | LSC · Random matrices, 99% sparse | compression ratio: 66.05x -> 69.22x | Helped |
| increase block size from 3x3 to 9x9 (ablation) | LSC · Random matrices, 99% sparse | compression ratio: 66.05x -> 51.45x | Hurt |
| LSC vs NB at low sparsity (50%) | LSC · Random matrices, 50% sparse | compression ratio: 1.88x (Bitmask) vs 1.81x | Hurt |
| LSC vs best baseline (NB) at high sparsity | LSC · Random matrices, 99% sparse | compression ratio: 61.87x -> 66.10x | Helped |
Closest papers by meaning
- Computation and Memory-Efficient Model Compression with Gradient Reweighting (NeurIPS 2025 · similarity 0.91)
- DominoSearch: Find layer-wise fine-grained N:M sparse schemes from dense neural networks (NeurIPS 2021 · similarity 0.90)
- Encoding Weights of Irregular Sparsity for Fixed-to-Fixed Model Compression (ICLR 2022 · similarity 0.90)
- Structured Compression by Weight Encryption for Unstructured Pruning and Quantization (CVPR 2020 · similarity 0.90)
- “Learning-Compression” Algorithms for Neural Net Pruning (CVPR 2018 · similarity 0.90)
- Targeted Low-rank Refinement: Enhancing Sparse Language Models with Precision (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.