Research map / Architectures and efficiency / Pruning and efficient tuning
Storage Efficient and Dynamic Flexible Runtime Channel Pruning via Deep Reinforcement Learning
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 |
|---|---|---|---|
| learn static and runtime pruning jointly (trade-off pruner, Rr=0.5) instead of training them separately (ablation) | DRL runtime pruning · CIFAR-10, sparsity 0.5 | accuracy (%): 90.03 -> 91.228 | Helped |
| joint DRL framework versus combining FPGM static pruning with FBS runtime pruning | DRL runtime pruning · CIFAR-10, sparsity 0.5 | accuracy (%): 90.456 -> 91.228 | Helped |
| DRL-learned layer-wise sparsity ratios versus FBS | M-CifarNet · CIFAR-10, sparsity 0.5 | accuracy (%): 89.88 -> 91.425 | Helped |
| add static pruning component (Rr 1 -> 0.5) to cut parameter storage (ablation) | DRL runtime pruning · CIFAR-10, sparsity 0.5 | accuracy (%): 91.425 -> 91.228 | Hurt |
| add static pruning component (Rr 1 -> 0.5) at sparsity 0.7 (ablation) | DRL runtime pruning · CIFAR-10, sparsity 0.7 | accuracy (%): 93.184 -> 92.714 | Hurt |
| raise CNN/importance-predictor learning rate from 1e-3 to 1e-2 (ablation) | DRL runtime pruning · CIFAR-10 | accuracy (%): 91.23 -> 10.00 | Hurt |
| lower CNN learning rate from 1e-3 to 1e-4 (ablation) | DRL runtime pruning · CIFAR-10 | accuracy (%): 91.23 -> 90.67 | Hurt |
| change DRL agent learning rate from 1e-6 to 1e-5 or 1e-7 (ablation) | DRL runtime pruning · CIFAR-10 | accuracy (%): 91.23 -> 89.95 (1e-5), 90.79 (1e-7) | Hurt |
| ours (Rr=0.5) versus AMC and FBS on ResNet-18 | ResNet-18 · ImageNet | top-1 accuracy after pruning (%): 66.63 -> 68.73 (AMC); 68.17 -> 68.73 (FBS) | Helped |
| ours versus AMC on MobileNet | MobileNet · ImageNet | top-1 accuracy after pruning (%): 70.5 -> 70.6 | Helped |
Closest papers by meaning
- Runtime Neural Pruning (NeurIPS 2017 · similarity 0.91)
- Rethinking the Smaller-Norm-Less-Informative Assumption in Channel Pruning of Convolution Layers (ICLR 2018 · similarity 0.90)
- REPrune: Channel Pruning via Kernel Representative Selection (AAAI 2024 · similarity 0.89)
- BilevelPruning: Unified Dynamic and Static Channel Pruning for Convolutional Neural Networks (CVPR 2024 · similarity 0.88)
- Jointly Training and Pruning CNNs via Learnable Agent Guidance and Alignment (CVPR 2024 · similarity 0.88)
- UPDP: A Unified Progressive Depth Pruner for CNN and Vision Transformer (AAAI 2024 · similarity 0.86)
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.