Research map / Architectures and efficiency / RNNs and normalization
EvoGrad: Efficient Gradient-Based Meta-Learning and Hyperparameter Optimization
Ondrej Bohdal, Yongxin Yang, Timothy Hospedales
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 |
|---|---|---|---|
| meta-learn rotation with EvoGrad | LeNet · MNIST rotated 30 deg | test accuracy (%): 81.79 -> 98.11 | Improved |
| replace T1-T2 with EvoGrad in Meta-Weight-Net | ResNet-32 · CIFAR-10 40% noise | test accuracy (%): 87.54 -> 87.74 | Improved |
| replace T1-T2 with EvoGrad in Meta-Weight-Net | ResNet-32 · CIFAR-100 40% noise | test accuracy (%): 58.64 -> 57.44 | Hurt |
| replace T1-T2 with EvoGrad in Meta-Weight-Net | ResNet-32 · CIFAR-10 20% noise | test accuracy (%): 90.33 -> 89.86 | Hurt |
| replace T1-T2 with EvoGrad in Meta-Weight-Net | ResNet-32 · CIFAR-10 0% noise | test accuracy (%): 92.04 -> 92.02 | No clear effect |
| add Meta-Weight-Net reweighting under label noise | ResNet-32 · CIFAR-100 40% noise | test accuracy (%): 43.01 -> 57.44 | Improved |
| add Meta-Weight-Net reweighting without label noise | ResNet-32 · CIFAR-10 0% noise | test accuracy (%): 92.89 -> 92.02 | Hurt |
| replace T1-T2 with EvoGrad in MetaXL | MetaXL · WikiAnn NER | average F1 (%): 71.13 -> 71.98 | Improved |
| replace T1-T2 with EvoGrad in MetaXL (our run) | MetaXL · WikiAnn NER | average F1 (%): 69.59 -> 71.98 | Improved |
| replace T1-T2 with EvoGrad in LFT | RelationNet ResNet10 · CUB 5-way 1-shot | accuracy (%): 48.38 -> 47.39 | Hurt |
| replace T1-T2 with EvoGrad in LFT | RelationNet ResNet10 · Plantae 5-way 5-shot | accuracy (%): 50.39 -> 52.92 | Improved |
| scale feature extractor from ResNet10 to ResNet34 with EvoGrad | LFT RelationNet · Plantae 5-way 1-shot | accuracy (%): 36.00 -> 38.23 | Improved |
Closest papers by meaning
- EvoGrad: Evolutionary-Weighted Gradient and Hessian Learning for Black-Box Optimization (AAAI 2026 · similarity 0.90)
- On Modulating the Gradient for Meta-Learning (ECCV 2020 · similarity 0.86)
- MetaOptimize: A Framework for Optimizing Step Sizes and Other Meta-parameters (ICML 2025 · similarity 0.85)
- Memory-Reduced Meta-Learning with Guaranteed Convergence (AAAI 2025 · similarity 0.85)
- EVOS: Efficient Implicit Neural Training via EVOlutionary Selector (CVPR 2025 · similarity 0.85)
- MetaDiff: Meta-Learning with Conditional Diffusion for Few-Shot Learning (AAAI 2024 · similarity 0.85)
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.
