Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Meta Networks
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 MetaNet instead of the strongest reported baseline | Omniglot previous split · 5-way one-shot | accuracy: 98.4 -> 98.95 | Helped |
| use MetaNet instead of the strongest reported baseline | Omniglot previous split · 20-way one-shot | accuracy: 95.0 -> 97.0 | Helped |
| use MetaNet instead of Matching Nets | Mini-ImageNet test · 5-way one-shot | accuracy: 43.56 ± 0.84 -> 49.21 ± 0.96 | Helped |
| add task-level fast weights to the embedding function (ablation) | MetaNet on Omniglot previous split · 5-way one-shot | accuracy: 98.4 -> 98.95 | Helped |
| add task-level fast weights to the embedding function (ablation) | MetaNet on MNIST out-of-domain · 10-way one-shot | accuracy: 71.6% -> 74.8% | Helped |
| add task-level fast weights to the base learner (ablation) | MetaNet on Omniglot previous split · 5-way one-shot | accuracy: 98.95 -> 98.45 | Hurt |
| add task-level fast weights to the base learner (ablation) | MetaNet on MNIST out-of-domain · 10-way one-shot | accuracy: 74.8% -> 72.3% | Hurt |
| train on harder N-way tasks than the K-way evaluation task | MetaNet · Omniglot previous split, N-way train / K-way test | accuracy: 20-way train, 5-way test: 99.55; 5-way train, 5-way test: 98.95 | Helped |
| continue meta-weight training on MNIST after Omniglot | MetaNet · Omniglot test after MNIST training | accuracy difference: -1.7% to 1.24% | Mixed |
Closest papers by meaning
- Improving Generalization of Meta-Learning With Inverted Regularization at Inner-Level (CVPR 2023 · similarity 0.89)
- Graph Metanetworks for Processing Diverse Neural Architectures (ICLR 2024 · similarity 0.89)
- Dynamic Kernel Selection for Improved Generalization and Memory Efficiency in Meta-Learning (CVPR 2022 · similarity 0.88)
- Large-Scale Meta-Learning with Continual Trajectory Shifting (ICML 2021 · similarity 0.88)
- Learning To Learn by Jointly Optimizing Neural Architecture and Weights (CVPR 2022 · similarity 0.88)
- A Simple Neural Attentive Meta-Learner (ICLR 2018 · 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.