Research map / Architectures and efficiency / Pruning and efficient tuning
TETRIS: TilE-matching the TRemendous Irregular Sparsity
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 input/output channel reordering before block pruning (ablation) | block sparsity · VGG16 ImageNet, various block sizes | top-1 accuracy drop vs 1x1: 2.3% ~ 6.0% drop -> 0.1% ~ 2.4% drop | Helped |
| add reordering to block pruning on LeNet and VGG14 (ablation) | block sparsity · MNIST, CIFAR-10 | accuracy vs sparsity across block sizes | Helped |
| increase block size from 1x1 to 32x32 with reordering (ablation) | TETRIS on VGG16 · VGG16 ImageNet | speedup: accuracy 73.36% -> 71.004%; speedup 2.35x vs dense | Helped |
| increase block size beyond 32 (ablation) | TETRIS on VGG16 · VGG16 ImageNet | speedup | Hurt |
| optimize reordering (incremental G updates, row maxima) (ablation) | reordering algorithm · VGG16 FC1 layer | pruning time: more than 2 hours -> less than 1 minute | Helped |
| add runtime permutation of inputs/outputs | inference · VGG16 layers | computation time: about 4% of layer computation time | Hurt |
Closest papers by meaning
- Encoding Weights of Irregular Sparsity for Fixed-to-Fixed Model Compression (ICLR 2022 · similarity 0.89)
- “Learning-Compression” Algorithms for Neural Net Pruning (CVPR 2018 · similarity 0.86)
- The Unreasonable Effectiveness of Random Pruning: Return of the Most Naive Baseline for Sparse Training (ICLR 2022 · similarity 0.85)
- Sparsest Models Elude Pruning: An Exposé of Pruning’s Current Capabilities (ICML 2024 · similarity 0.85)
- Targeted Low-rank Refinement: Enhancing Sparse Language Models with Precision (ICML 2025 · similarity 0.84)
- SequentialAttention++ for Block Sparsification: Differentiable Pruning Meets Combinatorial Optimization (NeurIPS 2024 · similarity 0.83)
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.