Research map / Label-efficient and robust learning / Knowledge distillation
A Lazy Approach to Long-Horizon Gradient-Based Meta-Learning
Muhammad Abdullah Jamal, Liqiang Wang, Boqing Gong
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 |
|---|---|---|---|
| add lazy teacher leap with long inner horizon | MAML · MiniImageNet 5-way 5-shot | accuracy (%): 63.11 -> 64.13 | Mixed |
| add lazy teacher leap with long inner horizon | MAML · MiniImageNet 5-way 1-shot | accuracy (%): 48.70 -> 48.26 | No clear effect |
| add lazy teacher leap with long inner horizon | Reptile · MiniImageNet 5-way 1-shot | accuracy (%): 49.97 -> 51.50 | Improved |
| use last 3 checkpoints instead of 1 in the convex hull (ablation) | Lazy Reptile · MiniImageNet 5-way 5-shot | accuracy (%): 67.22 -> 68.77 | Improved |
| use last 3 checkpoints instead of 1 in the convex hull (ablation) | Lazy MAML · MiniImageNet 5-way 1-shot | accuracy (%): 48.26 -> 48.17 | No clear effect |
| add lazy teacher to MAML with more shots | MAML · MiniImageNet 5-way 50-shot | accuracy (%): 67.13 -> 78.54 | Improved |
| add lazy teacher to MAML with more shots | MAML · MiniImageNet 5-way 20-shot | accuracy (%): 66.18 -> 74.8 | Improved |
| add lazy teacher to MAML on non-i.i.d. tasks (no class shuffling) | MAML · Omniglot 20-way 1-shot (non-i.i.d.) | accuracy (%): 81.51 -> 86.35 | Improved |
| use Reptile-style update instead of MAML-style in lazy method (ablation) | Lazy MAML · MiniImageNet 5-way 50-shot | accuracy (%): 78.54 -> 76.17 | Hurt |
| add lazy teacher to two-component weighting | Two-component weighting · CIFAR-LT-100 (imbalance 200) | top-1 error (%): 60.69 -> 58.67 | Improved |
| add lazy teacher to two-component weighting | Two-component weighting · CIFAR-LT-100 (imbalance 20) | top-1 error (%): 44.38 -> 43.68 | Improved |
| add lazy teacher to meta-attack | Meta-attack · CIFAR10 untargeted attack | avg queries: 1,583 -> 1,061 | Improved |
Closest papers by meaning
- Meta-Learning without Memorization (ICLR 2020 · similarity 0.90)
- Large-Scale Meta-Learning with Continual Trajectory Shifting (ICML 2021 · similarity 0.90)
- Meta-Learning in Neural Networks: A Survey (TPAMI 2021 · similarity 0.89)
- Improving Generalization of Meta-Learning With Inverted Regularization at Inner-Level (CVPR 2023 · similarity 0.89)
- Bridging Multi-Task Learning and Meta-Learning: Towards Efficient Training and Effective Adaptation (ICML 2021 · similarity 0.89)
- Memory-Reduced Meta-Learning with Guaranteed Convergence (AAAI 2025 · 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.
