Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
The Role of Deconfounding in Meta-learning
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 |
|---|---|---|---|
| MAML-Dropout + MAML-Bins combined (ablation) | MAML · Sinusoid regression 10-shot | MSE: 0.16 to 0.06 | Helped |
| MAML-Dropout + MAML-Bins combined | MAML · Pose prediction 10-shot | MSE: 3.098 to 2.396 | Helped |
| MAML-Dropout + MAML-Bins combined | MAML · Omniglot 20-way 1-shot | accuracy: 87.40 to 92.89 | Helped |
| MAML-Dropout + MAML-Bins combined | MAML · MiniImagenet 5-way 1-shot | accuracy: 32.93 to 39.89 | Helped |
| MAML-Dropout only (rate 0.1) | MAML · Drug activity prediction, Group 1 | mean R2: 0.366 to 0.410 | Helped |
| Adding MAML-Dropout and MAML-Bins to MAML outperforms MetaMix | MAML · Omniglot 20-way 1-shot and MiniImagenet 5-way 1-shot | accuracy: Omniglot 92.89 vs 92.06; MiniImagenet 39.89 vs 39.26 | Helped |
| Adding MAML-Dropout and MAML-Bins across backbones | ANIL, MetaSGD, T-NET · Sinusoid regression 10-shot | MSE: ANIL 0.10 to 0.08; MetaSGD 0.14 to 0.04; T-NET 0.11 to 0.07 | Helped |
| Adding MAML-Bins alone or MAML-Dropout alone on top of MAML (ablation) | MAML · Sinusoid regression, pose prediction, MiniImagenet, Omniglot | MSE / accuracy: both individually improve over MAML; combined is best | Helped |
| MAML-Bins added to drug activity prediction | MAML · Drug activity prediction | mean R2 | No clear effect |
| Dropout only on query sets during meta-training (MAML-Dropout), not on support sets | MAML · Drug activity prediction | mean R2: Group 1 0.366 to 0.410; Group 4 0.348 to 0.370 | Helped |
| Regularizer-based baselines (e.g., MR-MAML, IFSL) vs proposed method on sinusoid | MAML · Sinusoid regression 5-shot | MSE: IFSL 0.59 and MR-MAML 0.57 vs MAML 0.59 | Mixed |
| Adding MAML-Dropout and MAML-Bins to ANIL on pose prediction | ANIL · Pose prediction 10-shot | MSE: 6.746 to 6.289 | Helped |
Closest papers by meaning
- Meta-Learning Requires Meta-Augmentation (NeurIPS 2020 · similarity 0.89)
- Meta-Causal Learning for Single Domain Generalization (CVPR 2023 · similarity 0.87)
- A Meta-Learning Approach to Bayesian Causal Discovery (ICLR 2025 · similarity 0.86)
- Any-Way Meta Learning (AAAI 2024 · similarity 0.86)
- More Flexible PAC-Bayesian Meta-Learning by Learning Learning Algorithms (ICML 2024 · similarity 0.85)
- An Information-Theoretic Analysis of In-Context Learning (ICML 2024 · similarity 0.85)
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.