Research map / Optimization / Optimizers
Gradient-Based Meta-Learning with Learned Layerwise Metric and Subspace
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 a learned T transformation and gradient mask | MAML-style sine-wave regression · Sine wave regression, 5-shot | MSE: 1.07 $\pm$ 0.11 -> 0.76 $\pm$ 0.09 | Helped |
| add a learned T transformation and gradient mask | MAML-style sine-wave regression · Sine wave regression, 10-shot | MSE: 0.71 $\pm$ 0.07 -> 0.49 $\pm$ 0.05 | Helped |
| add a learned T transformation and gradient mask | MAML-style sine-wave regression · Sine wave regression, 20-shot | MSE: 0.50 $\pm$ 0.05 -> 0.33 $\pm$ 0.04 | Helped |
| add a learned gradient mask (ablation) | T-net · Sine wave regression, 5-shot | MSE: 0.83 $\pm$ 0.08 -> 0.76 $\pm$ 0.09 | Helped |
| add a learned gradient mask (ablation) | T-net · Sine wave regression, 10-shot | MSE: 0.56 $\pm$ 0.06 -> 0.49 $\pm$ 0.05 | Helped |
| add a learned gradient mask (ablation) | T-net · Sine wave regression, 20-shot | MSE: 0.38 $\pm$ 0.04 -> 0.33 $\pm$ 0.04 | Helped |
| add a T transformation (ablation) | M-net · Sine wave regression, 10-shot | MSE: 0.60 $\pm$ 0.06 -> 0.49 $\pm$ 0.05 | Helped |
| share mask parameters across weights contributing to the same activation instead of using per-weight masks (ablation) | MT-net · Sine wave regression, 10-shot | MSE: 0.51 $\pm$ 0.05 -> 0.49 $\pm$ 0.05 | Helped |
| use a learned T transformation | MAML · 10-shot sine wave regression across tested step sizes | MSE: 171.92 $\pm$ 25.04 -> 4.08 $\pm$ 0.30 | Helped |
| add the learned gradient mask to T-net (ablation) | T-net · 10-shot sine wave regression, α=1 | MSE: 4.15 $\pm$ 0.30 -> 0.61 $\pm$ 0.07 | Helped |
| use MT-net instead of T-net | Few-shot classification · Omniglot, 5-way 1-shot | accuracy: 99.4 $\pm$ 0.3 -> 99.5 $\pm$ 0.3 | No clear effect |
| use MT-net instead of T-net | Few-shot classification · Omniglot, 20-way 1-shot | accuracy: 96.1 $\pm$ 0.3 -> 96.2 $\pm$ 0.4 | No clear effect |
Closest papers by meaning
- Subspace Learning for Effective Meta-Learning (ICML 2022 · similarity 0.89)
- Meta-Learning and Universality: Deep Representations and Gradient Descent can Approximate any Learning Algorithm (ICLR 2018 · similarity 0.89)
- Memory-Reduced Meta-Learning with Guaranteed Convergence (AAAI 2025 · similarity 0.88)
- A Sample Complexity Separation between Non-Convex and Convex Meta-Learning (ICML 2020 · similarity 0.88)
- Optimistic Meta-Gradients (NeurIPS 2023 · similarity 0.88)
- Continuous-Time Meta-Learning with Forward Mode Differentiation (ICLR 2022 · 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.