Research map / Optimization / Optimizers
Meta-Learning with Warped Gradient Descent
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 WarpGrad preconditioning to MAML | MAML · miniImageNet 5-way 1-shot | accuracy: 48.7 -> 52.3 | Helped |
| add WarpGrad preconditioning to MAML | MAML · tieredImageNet 5-way 5-shot | accuracy: 70.3 -> 74.1 | Helped |
| add WarpGrad preconditioning to Leap | Leap · tieredImageNet 10-way 640-shot | accuracy: 76.52 -> 80.4 | Helped |
| use Warp-Leap instead of Leap | Leap · Omniglot 20-way 100-shot | accuracy: 75.5 -> 83.6 | Helped |
| train warp parameters offline rather than online (ablation) | Warp-Leap · Omniglot 25 meta-training tasks | test accuracy: 76.3±2.1 -> 84.4±1.7 | Helped |
| use the approximate meta-objective instead of the full objective (ablation) | Warp-Leap · Omniglot 25 meta-training tasks | test accuracy: 84.4±1.7 -> 83.1±2.7 | Hurt |
| meta-learn the task adaptation learning rate (ablation) | Warp-Leap · Omniglot 25 meta-training tasks | test accuracy: 84.4±1.7 -> 83.1±3.3 | Hurt |
| replace 3x3 convolutional warp with channel-wise scaling (ablation) | Warp-Leap · Omniglot 25 meta-training tasks | test accuracy: 84.4±1.7 -> 77.5±1.8 | Hurt |
| add batch normalization, ReLU, and residual connections to the warp (ablation) | Warp-Leap · Omniglot 25 meta-training tasks | test accuracy: 84.4±1.7 -> 86.3±1.1 | Helped |
| use two convolutional warp layers with batch normalization and residual connections (ablation) | Warp-Leap · Omniglot 25 meta-training tasks | test accuracy: 84.4±1.7 -> 88.0±1.0 | Helped |
| add task-adapted FiLM embeddings to the two-layer warp (ablation) | Warp-Leap · Omniglot 25 meta-training tasks | test accuracy: 88.0±1.0 -> 88.1±1.0 | No clear effect |
| use nonlinear instead of linear warp layers (ablation) | WarpGrad · Omniglot, random initialization | accuracy: 68.0±4.4 -> 81.3±4.0 | Helped |
Closest papers by meaning
- Optimistic Meta-Gradients (NeurIPS 2023 · similarity 0.90)
- Large-Scale Meta-Learning with Continual Trajectory Shifting (ICML 2021 · similarity 0.88)
- Continuous-Time Meta-Learning with Forward Mode Differentiation (ICLR 2022 · similarity 0.88)
- 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 0.87)
- Provable Meta-Learning with Low-Rank Adaptations (NeurIPS 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.