Research map / Architectures and efficiency / Neural architecture search
Meta-Learning of Neural Architectures for Few-Shot Learning
Check 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 |
|---|---|---|---|
| Meta-learn and adapt both architecture and weights rather than using the REPTILE fixed architecture | MetaNAS vs REPTILE · MiniImagenet, 1-shot, 5-way | accuracy (%): 48.17 ± 0.93 -> 54.86 ± 0.77 | Helped |
| Meta-learn and adapt both architecture and weights rather than using the REPTILE fixed architecture | MetaNAS vs REPTILE · Omniglot, 1-shot, 20-way | accuracy (%): 89.33 ± 0.07 -> 94.40 ± 0.61 | Helped |
| Use MetaNAS instead of AutoMeta under the same REPTILE training pipeline | MetaNAS vs AutoMeta · Omniglot, 1-shot, 20-way | accuracy (%): 95.60 ± 0.12 -> 94.40 ± 0.61 | Hurt |
| Use MetaNAS instead of AutoMeta under the same REPTILE training pipeline | MetaNAS vs AutoMeta · MiniImagenet, 1-shot, 5-way | accuracy (%): 50.50 ± 0.76 -> 54.86 ± 0.77 | Helped |
| Scale up MetaNAS architecture, train longer, and add regularization | MetaNAS* · MiniImagenet, 1-shot, 5-way | accuracy (%): 57.6 ± 0.2 -> 63.1 ± 0.3 | Helped |
| Scale up MetaNAS architecture, train longer, and add regularization | MetaNAS* · MiniImagenet, 5-shot, 5-way | accuracy (%): 74.7 ± 0.2 -> 79.5 ± 0.2 | Helped |
| Scale up MetaNAS architecture, train longer, and add regularization | MetaNAS* · Omniglot, 1-shot, 20-way | accuracy (%): 97.65 ± 0.05 -> 97.87 ± 0.07 | Helped |
| Sparsify both operation mixtures and input-node combinations before pruning (ablation) | DARTS · CIFAR-10, after pruning | accuracy: 16.7 ± 2.1 % -> 84.7 ± 0.4 % | Helped |
| Also sparsify input nodes, beyond sparsifying operations only (ablation) | DARTS · CIFAR-10, after pruning | accuracy: 60.2 ± 4.0 % -> 84.7 ± 0.4 % | Helped |
| Sparsify operations only, leaving input nodes unsparsified (ablation) | DARTS · CIFAR-10, after pruning | accuracy: 16.7 ± 2.1 % -> 60.2 ± 4.0 % | Helped |
| Hard-prune without sparsifying operations or inputs (ablation) | DARTS · CIFAR-10, before vs after pruning | accuracy: 87.9 ± 0.2 % -> 16.7 ± 2.1 % | Hurt |
| Hard-prune after sparsifying operations only (ablation) | DARTS · CIFAR-10, before vs after pruning | accuracy: 84.4 ± 0.4 % -> 60.2 ± 4.0 % | Hurt |
Closest papers by meaning
- Rapid Model Architecture Adaption for Meta-Learning (NeurIPS 2022 · similarity 0.92)
- Towards Fast Adaptation of Neural Architectures with Meta Learning (ICLR 2020 · similarity 0.92)
- Global Convergence of MAML and Theory-Inspired Neural Architecture Search for Few-Shot Learning (CVPR 2022 · similarity 0.92)
- Revisiting Neural Networks for Few-Shot Learning: A Zero-Cost NAS Perspective (ICML 2025 · similarity 0.91)
- Neural Fine-Tuning Search for Few-Shot Learning (ICLR 2024 · similarity 0.90)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · similarity 0.89)
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.