Research map / Architectures and efficiency / Pruning and efficient tuning
Runtime Neural Pruning
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 |
|---|---|---|---|
| Runtime neural pruning (input-dependent channel-group pruning) instead of static filter pruning | VGG-16 on ILSVRC2012 · ILSVRC2012-val top-5 | top-5 error increase (%): at 5x: 3.58 vs 14.6 for filter pruning | Helped |
| Runtime neural pruning at larger speed-up | VGG-16 on ILSVRC2012 · ILSVRC2012-val top-5 | top-5 error increase (%): at 3x: 2.32 vs 3.2 for filter pruning, but comparable to Jaderberg (2.3) | Helped |
| Increase speed-up from 5x to 10x | RNP on VGG-16 · ILSVRC2012-val top-5 | top-5 error increase (%): 3.58 to 4.89 (+1.31) | Hurt |
| Runtime neural pruning on VGG-16 GPU inference | VGG-16 on ILSVRC2012-val · GPU inference time (Titan X, batch 64) | mean inference time (ms): 3.26 to 0.880 ms at 5x theoretical speed-up (3.7x actual) | Helped |
| Runtime neural pruning vs. vanilla narrower network on CIFAR-10 and CIFAR-100 | 4-layer CNN (CIFAR-10) and VGG-16 (CIFAR-100) · CIFAR-10 / CIFAR-100 | accuracy vs multiplications: RNP consistently outperforms vanilla channel reduction by a large margin | Helped |
| Adding a penalty-based pruning policy on LFW-T 3-class task | 3-layer CNN on LFW-T · LFW-T test accuracy | accuracy (%) and multiplications: 91.1% original to 91.75% with 2x fewer multiplications | Helped |
Closest papers by meaning
- Storage Efficient and Dynamic Flexible Runtime Channel Pruning via Deep Reinforcement Learning (NeurIPS 2020 · similarity 0.91)
- Pruning from Scratch (AAAI 2020 · similarity 0.90)
- Structured Pruning of Neural Networks With Budget-Aware Regularization (CVPR 2019 · similarity 0.90)
- Dynamic Structure Pruning for Compressing CNNs (AAAI 2023 · similarity 0.90)
- Jointly Training and Pruning CNNs via Learnable Agent Guidance and Alignment (CVPR 2024 · similarity 0.89)
- BilevelPruning: Unified Dynamic and Static Channel Pruning for Convolutional Neural Networks (CVPR 2024 · 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.