Research map / Optimization / Optimizers
On Modulating the Gradient for Meta-Learning
Christian Simon, Piotr Koniusz, Richard Nock, Mehrtash Harandi
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 |
|---|---|---|---|
| replace CAVIA-style methods with ModGrad modulation (1-step Conv-4) | CAVIA (512) · mini-ImageNet 5-way 1-shot | accuracy: 51.82 -> 53.20 | Improved |
| replace CAVIA-style methods with ModGrad modulation (1-step Conv-4) | CAVIA (512) · mini-ImageNet 5-way 5-shot | accuracy: 65.85 -> 69.17 | Improved |
| replace Meta-Curvature with ModGrad on WRN-28-10 | Meta-Curvature · mini-ImageNet 5-way 1-shot | accuracy: 64.40 -> 65.72 | Improved |
| replace Meta-Curvature with ModGrad on WRN-28-10 | Meta-Curvature · mini-ImageNet 5-way 5-shot | accuracy: 80.21 -> 81.17 | Improved |
| replace LEAP with ModGrad on multi-shot Omniglot | LEAP · Omniglot 20-way multi-shot | accuracy: 83.3 -> 85.1 | Improved |
| use ModGrad instead of CAVIA on image completion (10 random pixels) | CAVIA · CelebA image completion (10 random pixels) | accuracy: 0.037 -> 0.034 | Improved |
| run more inner-loop steps (5 instead of 1) in ModGrad (ablation) | ModGrad · mini-ImageNet 5-way 5-shot | accuracy: 69.17 -> 68.24 | No clear effect |
| run more inner-loop steps (5 instead of 1) in ModGrad (ablation) | ModGrad · mini-ImageNet 5-way 1-shot | accuracy: 53.20 -> 52.76 | No clear effect |
| increase column dimension u from 5 to 10 (ablation) | ModGrad · mini-ImageNet 5-way 5-shot | accuracy: 69.17 -> 67.62 | Hurt |
| decrease column dimension u from 5 to 1 (ablation) | ModGrad · mini-ImageNet 5-way 1-shot | accuracy: 53.20 -> 53.13 | No clear effect |
| use ModGrad with 1 step instead of MAML with 30 steps on ResNet-34 20-shot | MAML · mini-ImageNet 5-way 20-shot | accuracy: 2% | Improved |
| add ModGrad under gradient noise | Meta-SGD, CAVIA, T-Net, MAML · mini-ImageNet (noisy gradients) | accuracy: 10% | Improved |
Closest papers by meaning
- MetaGrad: Adaptation using Multiple Learning Rates in Online Learning (JMLR 2021 · similarity 0.89)
- Meta-Learning in Neural Networks: A Survey (TPAMI 2021 · similarity 0.89)
- Optimistic Meta-Gradients (NeurIPS 2023 · similarity 0.89)
- Memory-Reduced Meta-Learning with Guaranteed Convergence (AAAI 2025 · similarity 0.88)
- MetaDiff: Meta-Learning with Conditional Diffusion for Few-Shot Learning (AAAI 2024 · similarity 0.88)
- On the Stability and Generalization of Meta-Learning (NeurIPS 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.
