Research map / Image generation / Image restoration
Supplementary Meta-Learning: Towards a Dynamic Model for Deep Neural Networks
Feihu Zhang, Benjamin W. Wah
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 dynamic MLNN layers to the CNN | Arc 1 super-resolution · Super-resolution Dataset-5 | PSNR (dB): 32.64 -> 32.87 | Improved |
| add MLNN to Arc 1 on less diverse data | Arc 1 super-resolution · Super-resolution Dataset-1 | PSNR (dB): 29.97 -> 30.11 | Improved |
| add MLNN to 21-layer residual net | Arc 4 super-resolution · Super-resolution Dataset-5 | PSNR (dB): 32.99 -> 33.11 | Improved |
| add MLNN to denoising networks | Arc 1 denoising · Denoising BSD | PSNR (dB): 28.71 -> 28.87 | Improved |
| add dynamic dot-product and addition rather than concatenation of the same branches (ablation) | MLNN Arc 1 · Super-resolution Dataset-5 | PSNR (dB): 32.71 -> 32.87 | Improved |
| add dynamic Fw branch instead of fixed weights (ablation) | MLNN Arc 1 · Super-resolution upsampling | PSNR (dB): 32.68 -> 32.87 | Improved |
| add learned Fb instead of identity mapping (ablation) | MLNN Arc 1 · Super-resolution upsampling | PSNR (dB): 32.7 -> 32.87 | Improved |
| add dynamic Fw branch instead of fixed weights (ablation) | MLNN AlexNet · ImageNet | top-1 error: top-1 error rate increases by 0.61% | Improved |
| add learned Fb instead of identity mapping (ablation) | MLNN AlexNet · ImageNet | top-1 error: top-1 error rate increases by 0.73% | Improved |
| add MLNN to more AlexNet layers (MLNN-1 to MLNN-3) (ablation) | AlexNet · ImageNet | top-1 error (%): 42.2 -> 41.1 | Improved |
| convert further layers to MLNN (MLNN-5) (ablation) | AlexNet · ImageNet | accuracy: accuracy begins to drop by 0.2% | Hurt |
| add MLNN to MNIST fully-connected layer | Arc 5 · MNIST | error rate (%): 1.11 -> 0.71 | Improved |
Closest papers by meaning
- Learning to Learn Parameterized Classification Networks for Scalable Input Images (ECCV 2020 · similarity 0.90)
- Deep Neural Network Parameter Selection via Dataset Similarity Under Meta-Learning Framework (TPAMI 2025 · similarity 0.90)
- Meta Networks (ICML 2017 · similarity 0.89)
- Meta Convolutional Neural Networks for Single Domain Generalization (CVPR 2022 · similarity 0.88)
- Dynamic Kernel Selection for Improved Generalization and Memory Efficiency in Meta-Learning (CVPR 2022 · similarity 0.88)
- LAuReL: Learned Augmented Residual Layer (ICML 2025 · 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.
