Research map / Graph learning / Knowledge and temporal graphs
Automated Relational 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 the ARML relational meta-knowledge graph rather than HSML's hierarchical structure | few-shot classification · Plain-Multi, 5-way 5-shot, Bird | accuracy: 71.68± 0.73% -> 73.34±0.70 | Helped |
| use the ARML relational meta-knowledge graph rather than HSML's hierarchical structure | few-shot classification · Art-Multi, 5-way 5-shot, Avg. Pencil | accuracy: 53.93± 0.72% -> 55.27±0.73% | Helped |
| use ARML instead of MetaSGD | few-shot classification · miniImageNet, 5-way 1-shot | accuracy: 50.47± 1.87% -> 50.42± 1.73% | No clear effect |
| add prototype-based relational graph rather than using mean pooling over samples (ablation) | ARML · Plain-Multi, 5-way 5-shot, Bird | accuracy: 71.96±0.72% -> 73.34±0.70% | Helped |
| represent task samples with prototypes instead of a sample-level relational graph (ablation) | ARML · Plain-Multi, 5-way 5-shot, Bird | accuracy: 72.86±0.74% -> 73.34±0.70% | Helped |
| add links between prototypes (ablation) | ARML · Plain-Multi, 5-way 5-shot, Bird | accuracy: 72.53±0.72% -> 73.34±0.70% | Helped |
| add the meta-knowledge graph (ablation) | ARML · Plain-Multi, 5-way 5-shot, Bird | accuracy: 71.23±0.75% -> 73.34±0.70% | Helped |
| add reconstruction loss (ablation) | ARML · Plain-Multi, 5-way 5-shot, Bird | accuracy: 70.99±0.74% -> 73.34±0.70% | Helped |
| use the selected encoder/decoder instead of MLP (ablation) | ARML · Plain-Multi, 5-way 5-shot, Bird | accuracy: 72.36±0.72% -> 73.34±0.70% | Helped |
| use a customized gate for each parameter instead of sharing the gate within convolutional filters (ablation) | ARML · Plain-Multi, 5-way 5-shot, Bird | accuracy: 72.83±0.72% -> 73.34±0.70% | Helped |
| replace sigmoid modulation with tanh (ablation) | ARML · Plain-Multi, 5-way 5-shot, Bird | accuracy: 73.34±0.70% -> 73.45±0.71% | No clear effect |
| replace sigmoid modulation with FiLM (ablation) | ARML · Plain-Multi, 5-way 5-shot, Bird | accuracy: 73.34±0.70% -> 72.95±0.73% | Hurt |
Closest papers by meaning
- MoEMeta: Mixture-of-Experts Meta Learning for Few-Shot Relational Learning (NeurIPS 2025 · similarity 0.90)
- Hierarchically Structured Meta-learning (ICML 2019 · similarity 0.90)
- Online Structured Meta-learning (NeurIPS 2020 · similarity 0.88)
- Bridging Multi-Task Learning and Meta-Learning: Towards Efficient Training and Effective Adaptation (ICML 2021 · similarity 0.88)
- Task Groupings Regularization: Data-Free Meta-Learning with Heterogeneous Pre-trained Models (ICML 2024 · similarity 0.87)
- Meta-Learning without Memorization (ICLR 2020 · 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.