Research map / Optimization / Optimizers
Adaptive Gradient-Based Meta-Learning Methods
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 |
|---|---|---|---|
| replace fixed-rate Adam meta-optimizer with ARUBA per-coordinate learning rate | Reptile · 5-way Mini-ImageNet 1-shot | accuracy (%): 49.97 -> 50.73 | Helped |
| replace Adam with ARUBA per-coordinate learning rate | Reptile · 5-way Mini-ImageNet 5-shot | accuracy (%): 65.99 -> 65.69 | No clear effect |
| replace Adam with ARUBA per-coordinate learning rate | Reptile · 20-way Omniglot 1-shot | accuracy (%): 89.43 -> 86.67 | Hurt |
| add meta-test-time learning-rate updating (ARUBA++) to ARUBA | Reptile+ARUBA · 20-way Omniglot 1-shot | accuracy (%): 86.67 -> 89.66 | Helped |
| add meta-test-time learning-rate updating (ARUBA++) to ARUBA | Reptile+ARUBA · 20-way Omniglot 5-shot | accuracy (%): 96.61 -> 97.49 | Helped |
| add meta-test-time learning-rate updating (ARUBA++) to ARUBA | Reptile+ARUBA · 5-way Mini-ImageNet 1-shot | accuracy (%): 50.73 -> 50.35 | No clear effect |
| learn per-coordinate learning rate (Meta-SGD) on MAML | 2nd-order MAML · 20-way Omniglot 1-shot | accuracy (%): 95.8 -> 95.93 | Helped |
| learn per-coordinate learning rate (Meta-SGD) on MAML | 2nd-order MAML · 5-way Mini-ImageNet 5-shot | accuracy (%): 63.11 -> 64.03 | Helped |
| apply ARUBA learning-rate adaptation to multi-step MAML | MAML · few-shot benchmarks | accuracy (%) | No clear effect |
Closest papers by meaning
- Provable Guarantees for Gradient-Based Meta-Learning (ICML 2019 · similarity 0.94)
- A Sample Complexity Separation between Non-Convex and Convex Meta-Learning (ICML 2020 · similarity 0.90)
- Optimistic Meta-Gradients (NeurIPS 2023 · similarity 0.90)
- On the Stability and Generalization of Meta-Learning (NeurIPS 2024 · similarity 0.89)
- Memory-Reduced Meta-Learning with Guaranteed Convergence (AAAI 2025 · similarity 0.89)
- Theoretical Investigations and Practical Enhancements on Tail Task Risk Minimization in Meta Learning (NeurIPS 2024 · similarity 0.88)
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.