Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Hierarchically Structured 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 |
|---|---|---|---|
| use hierarchical task clustering and cluster-specific initialization | MAML · Toy regression, 5-shot | MSE: 2.205\pm0.121 -> 0.856\pm 0.073 | Helped |
| use hierarchical task clustering and cluster-specific initialization | MAML · Toy regression, 10-shot | MSE: 0.761\pm0.068 -> 0.161\pm 0.021 | Helped |
| use hierarchical task clustering and cluster-specific initialization | MUMOMAML · Four-dataset few-shot classification, 5-way 1-shot average | accuracy: 46.50\% -> 49.35\% | Helped |
| use hierarchical task clustering and cluster-specific initialization | MUMOMAML · Four-dataset few-shot classification, 5-way 5-shot average | accuracy: 59.41\% -> 62.39\% | Helped |
| use HSML on tasks without obvious heterogeneity | MAML · MiniImagenet, 5-way 1-shot | accuracy: 48.70\pm1.84\% -> 50.38\pm 1.85\% | No clear effect |
| dynamically add clusters as task groups change | HSML-Static (10C) · Continual few-shot classification, 5-way 1-shot average | accuracy: 46.63\% -> 47.96\% | Helped |
| reduce the cluster configuration from (4,2,1) to (2,2,1) (ablation) | HSML · Few-shot classification, Bird, 5-way 1-shot | accuracy: 60.98\pm1.50\% -> 58.37\pm 1.42\% | Hurt |
| replace recurrent autoencoder aggregator with mean pooling autoencoder aggregator (ablation) | HSML · Four-dataset few-shot classification, 5-way 1-shot average | accuracy: 49.35\% -> 46.14\% | Hurt |
| replace recurrent autoencoder aggregator with mean pooling autoencoder aggregator (ablation) | HSML · Four-dataset few-shot classification, 5-way 3-shot average | accuracy: 58.80\% -> 58.07\% | Hurt |
| replace recurrent autoencoder aggregator with mean pooling autoencoder aggregator (ablation) | HSML · Four-dataset few-shot classification, 5-way 5-shot average | accuracy: 62.39\% -> 61.40\% | Hurt |
| replace recurrent autoencoder aggregator with mean pooling autoencoder aggregator (ablation) | HSML · Four-dataset few-shot classification, 5-way 8-shot average | accuracy: 65.01\% -> 64.97\% | No clear effect |
Closest papers by meaning
- Automated Relational Meta-learning (ICLR 2020 · similarity 0.90)
- Online Structured Meta-learning (NeurIPS 2020 · similarity 0.90)
- Hierarchical Gaussian Mixture based Task Generative Model for Robust Meta-Learning (NeurIPS 2023 · similarity 0.89)
- Structured Prediction for Conditional Meta-Learning (NeurIPS 2020 · similarity 0.89)
- Task Groupings Regularization: Data-Free Meta-Learning with Heterogeneous Pre-trained Models (ICML 2024 · similarity 0.87)
- Learning to Learn with Contrastive Meta-Objective (NeurIPS 2025 · similarity 0.86)
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.