Research map / Architectures and efficiency / Neural architecture search
DHP: Differentiable Meta Pruning via HyperNetworks
Yawei Li, Shuhang Gu, Kai Zhang, Luc Van Gool, Radu Timofte
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 latent vector sharing among correlated layers (lambda 3^-4, 50% FLOPs) (ablation) | DHP ResNet-56 · CIFAR10 | Top-1 error (%): 7.11 -> 6.52 | Improved |
| add latent vector sharing (lambda 3^-4, 38% FLOPs) (ablation) | DHP ResNet-56 · CIFAR10 | Top-1 error (%): 7.98 -> 6.86 | Improved |
| add latent vector sharing (lambda 2^-4, 50% FLOPs) (ablation) | DHP ResNet-56 · CIFAR10 | Top-1 error (%): 6.85 -> 7.13 | Hurt |
| use l1 instead of l2 sparsity regularizer (ablation) | DHP ResNet-20 · CIFAR10 | Top-1 error (%): 8.66 -> 8.46 | Improved |
| use l1 instead of l2 sparsity regularizer (ablation) | DHP ResNet-164 · CIFAR10 | Top-1 error (%): 5.18 -> 5.22 | No clear effect |
| prune widened MobileNetV1 (alpha 2) with DHP instead of using original | MobileNetV1 · Tiny-ImageNet | Top-1 error (%): 52.71 -> 50.75 | Improved |
| prune widened MobileNetV2 (alpha 2) with DHP instead of using original | MobileNetV2 · Tiny-ImageNet | Top-1 error (%): 44.75 -> 43.82 | Improved |
| use DHP-50 instead of uniform scaling MobileNetV1-0.75 | MobileNetV1 · Tiny-ImageNet | Top-1 error (%): 54.22 -> 51.63 | Improved |
| use DHP-10 instead of uniform scaling MobileNetV2-0.3 | MobileNetV2 · Tiny-ImageNet | Top-1 error (%): 53.99 -> 52.43 | Improved |
| use DHP instead of MetaPruning | MobileNetV2 (11% FLOPs) · Tiny-ImageNet | Top-1 error (%): 56.72 -> 52.43 | Improved |
| use DHP-60 instead of baseline for SRResNet | SRResNet · Urban100 x4 SR | PSNR (dB): 25.88 -> 25.76 | Hurt |
| use DHP-20 instead of baseline for DnCNN | DnCNN · BSD68 denoising | PSNR (dB): 24.93 -> 24.84 | Hurt |
Closest papers by meaning
- DMCP: Differentiable Markov Channel Pruning for Neural Networks (CVPR 2020 · similarity 0.90)
- AutoPrune: Automatic Network Pruning by Regularizing Auxiliary Parameters (NeurIPS 2019 · similarity 0.90)
- Pruning from Scratch (AAAI 2020 · similarity 0.89)
- SequentialAttention++ for Block Sparsification: Differentiable Pruning Meets Combinatorial Optimization (NeurIPS 2024 · similarity 0.88)
- MDP: Multidimensional Vision Model Pruning with Latency Constraint (CVPR 2025 · similarity 0.88)
- Lightweight Image Super-Resolution via Flexible Meta Pruning (ICML 2024 · similarity 0.87)
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.
