Research map / Architectures and efficiency / Pruning and efficient tuning
Parameter-Efficient Masking Networks
Yue Bai, Huan Wang, Xu Ma, Yitian Zhang, ZHIQIANG TAO, Yun Fu
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 |
|---|---|---|---|
| reduce unique parameters from MP to RP 1e-5 (ablation) | PEMN ConvMixer 6-block dim 256 · CIFAR10 | test accuracy: 88.97 (0.08) -> 81.01 (0.38) | Hurt |
| reduce unique parameters from MP to RP 1e-5 (ablation) | PEMN ConvMixer 8-block dim 512 · CIFAR10 | test accuracy: 93.21 (0.05) -> 87.40 (0.22) | Hurt |
| reduce unique parameters from MP to RP 1e-1/1e-2/1e-3 (ablation) | PEMN ConvMixer · CIFAR10 | test accuracy: 88.97 (0.08) -> 88.89 (0.13) | No clear effect |
| reduce unique parameters from MP to RP 1e-4/1e-5 (ablation) | PEMN ViT 6-block dim 256 · CIFAR10 | test accuracy: 76.70 (0.10) -> 64.84 (0.21) | Hurt |
| reduce unique parameters from MP to RP 1e-5 (ablation) | PEMN ViT 8-block dim 512 · CIFAR10 | test accuracy: 83.66 (0.12) -> 78.84 (0.29) | Hurt |
| increase hidden dimension from 256 to 512 (ablation) | PEMN MP ConvMixer 6-block · CIFAR10 | test accuracy: 88.97 (0.08) -> 92.02 (0.02) | Improved |
| increase depth from 6 to 8 blocks (ablation) | PEMN MP ConvMixer dim 512 · CIFAR10 | test accuracy: 92.02 (0.02) -> 93.21 (0.05) | Improved |
| increase depth from 6 to 8 blocks (ablation) | PEMN MP ViT dim 256 · CIFAR10 | test accuracy: 76.70 (0.10) -> 79.23 (0.22) | Improved |
| fix random weights and learn only a mask (Mask) versus dense training | sparse selection on random weights · CIFAR10 | test accuracy | Hurt |
| use PEMN MP instead of 90% sparse training baseline | ResNet56 · CIFAR10 | test accuracy: 86.35 (0.22) -> 88.13 (0.19) | Improved |
| use PEMN MP instead of 90% sparse training baseline | ResNet32 · CIFAR100 | test accuracy: 49.21 (0.29) -> 51.76 (0.25) | Improved |
| use PEMN MP instead of 90% sparse training baseline | ResNet56 · CIFAR100 | test accuracy: 55.21 (0.36) -> 56.39 (0.29) | Improved |
Closest papers by meaning
- Compressing Neural Networks: Towards Determining the Optimal Layer-wise Decomposition (NeurIPS 2021 · similarity 0.87)
- Accelerated Sparse Neural Training: A Provable and Efficient Method to Find N:M Transposable Masks (NeurIPS 2021 · similarity 0.87)
- Weight Fixing Networks (ECCV 2022 · similarity 0.87)
- Learning Compact Representations of Neural Networks using DiscriminAtive Masking (DAM) (NeurIPS 2021 · similarity 0.87)
- Scalable Model Compression by Entropy Penalized Reparameterization (ICLR 2020 · similarity 0.86)
- Auto-Compressing Networks (NeurIPS 2025 · 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.
