Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
A Simple Neural Attentive Meta-Learner
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 |
|---|---|---|---|
| remove temporal convolution layers (attention only) (ablation) | SNAIL · mini-ImageNet 5-way 1-shot | accuracy: 55.71% -> 49.9% | Hurt |
| remove attention (TC only) (ablation) | SNAIL · mini-ImageNet 5-way 5-shot | accuracy: 68.88% -> 61.2% | Hurt |
| remove temporal convolutions on Omniglot (ablation) | SNAIL · Omniglot | accuracy | No clear effect |
| use shallow mini-ImageNet embedding (ablation) | SNAIL · mini-ImageNet 5-way 1-shot | accuracy: 55.71% -> 45.1% | Hurt |
| give MAML SNAIL's deeper embedding (ablation) | MAML · mini-ImageNet 5-way 1-shot | accuracy: 30.1% & 75.2% | Hurt |
| replace SNAIL with stacked LSTM (ablation) | SNAIL · Omniglot 5-way | accuracy: 99.07% -> 78.1% | Hurt |
| TC-only SNAIL in tabular MDPs (ablation) | SNAIL · MDP N=100 | normalized reward: 0.941 -> 0.728 | Hurt |
| TC-only SNAIL in bandits (ablation) | SNAIL · multi-armed bandits | mean reward | No clear effect |
| attention-only (Transformer-like) SNAIL in RL (ablation) | SNAIL · bandits and MDPs | reward | Hurt |
| random architecture permutations of SNAIL (ablation) | SNAIL · Omniglot 5-way | accuracy: 98.62% ± 0.13% and 99.71% ± 0.08% | No clear effect |
| freeze Omniglot SNAIL, relearn mini-ImageNet embedding (ablation) | SNAIL · mini-ImageNet | accuracy: 50.62% and 62.34% | Mixed |
| freeze mini-ImageNet SNAIL, relearn Omniglot embedding (ablation) | SNAIL · Omniglot 5-way | accuracy: 98.66% and 99.56% | No clear effect |
Closest papers by meaning
- Dynamic Kernel Selection for Improved Generalization and Memory Efficiency in Meta-Learning (CVPR 2022 · similarity 0.88)
- Meta Networks (ICML 2017 · similarity 0.88)
- Learning Large-scale Neural Fields via Context Pruned Meta-Learning (NeurIPS 2023 · similarity 0.88)
- Memory Efficient Meta-Learning with Large Images (NeurIPS 2021 · similarity 0.87)
- First-order ANIL provably learns representations despite overparametrisation (ICLR 2024 · similarity 0.86)
- Unsupervised Meta-Learning via In-Context Learning (ICLR 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.