Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Improving Generalization in Meta-learning via Task Augmentation
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 MetaMix to ANIL++ | ANIL++ · Drug activity prediction, Groups 1–4 | mean R²: 0.367 -> 0.413; 0.315 -> 0.337; 0.335 -> 0.381; 0.362 -> 0.380 | Helped |
| Add MetaMix to MetaSGD | MetaSGD · Pose prediction, 10-shot | MSE: 2.803±0.239 -> 2.390±0.191 | Helped |
| Add MetaMix to MetaSGD | MetaSGD · Pose prediction, 15-shot | MSE: 2.331±0.182 -> 1.952±0.154 | Helped |
| Add MMCF to MAML | MAML · MiniImagenet, 5-way 1-shot | accuracy: 32.93±1.70% -> 39.26 ± 1.79% | Helped |
| Add Channel Shuffle to MetaMix (ablation) | MAML-MetaMix · Omniglot, 20-way 1-shot | accuracy: 91.53±0.53% -> 92.06 ± 0.51% | Helped |
| Add Channel Shuffle to MetaMix (ablation) | MAML-MetaMix · MiniImagenet, 5-way 1-shot | accuracy: 38.53±1.79% -> 39.26 ± 1.79% | Helped |
| Replace query-only training with support-query MetaMix (ablation) | MAML · Omniglot, 20-way 1-shot | accuracy: 90.65±0.56% -> 91.53±0.53% | Helped |
| Replace query-only training with support-query MetaMix (ablation) | MAML · MiniImagenet, 5-way 1-shot | accuracy: 34.56±1.77% -> 38.53±1.79% | Helped |
| Mix support samples with support samples rather than support-query samples (ablation) | ANIL++ · Drug activity prediction, Group 1 | mean R²: 0.413 -> 0.224 | Hurt |
| Replace support-query MetaMix with support-query concatenation (ablation) | ANIL++ · Drug activity prediction, Group 1 | mean R²: 0.413 -> 0.376 | Hurt |
Closest papers by meaning
- Improving Generalization of Meta-Learning With Inverted Regularization at Inner-Level (CVPR 2023 · similarity 0.92)
- Meta-Learning Requires Meta-Augmentation (NeurIPS 2020 · similarity 0.91)
- Efficient Heterogeneous Meta-Learning via Channel Shuffling Modulation (ICLR 2024 · similarity 0.90)
- Set-based Meta-Interpolation for Few-Task Meta-Learning (NeurIPS 2022 · similarity 0.89)
- Task Groupings Regularization: Data-Free Meta-Learning with Heterogeneous Pre-trained Models (ICML 2024 · similarity 0.88)
- Unsupervised Meta-Learning via In-Context Learning (ICLR 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.