Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Unsupervised Meta-Learning for Few-Shot Image Classification
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 |
|---|---|---|---|
| generate synthetic meta-training tasks by random sampling and augmenting images, without labels | UMTRA · Omniglot and Mini-ImageNet few-shot classification | accuracy | Helped |
| use UMTRA instead of CACTUs-MAML with ACAI/DC | few-shot classification · Omniglot 5-way 1-shot, 5-shot, 20-way 1-shot and 5-shot | accuracy: 83.80 vs 68.84; 95.43 vs 87.78; 74.25 vs 48.09; 92.12 vs 73.36 | Helped |
| use UMTRA instead of CACTUs-MAML with ACAI/DC | few-shot classification · Mini-ImageNet 5-way 1-shot | accuracy: 39.93 vs 39.90 | Helped |
| use UMTRA instead of CACTUs-MAML with ACAI/DC | few-shot classification · Mini-ImageNet 5-way 5-, 20-, and 50-shot | accuracy: 50.73 vs 53.97; 61.11 vs 63.84; 67.15 vs 69.64 | Hurt |
| use UMTRA instead of supervised MAML | few-shot classification · Omniglot and Mini-ImageNet | accuracy: Omniglot 5-way 1-shot: 83.80 vs 94.46; 5-way 5-shot: 95.43 vs 98.83 | Hurt |
| use randomly zeroed pixels instead of identity augmentation (ablation) | UMTRA · Omniglot 5-way 1-shot | accuracy: 52.93 -> 56.23 | Helped |
| tune hyperparameters for randomly zeroed pixel augmentation (ablation) | UMTRA · Omniglot 5-way 1-shot | accuracy: 56.23 -> 67.00 | Helped |
| add random shift to randomly zeroed pixels (ablation) | UMTRA · Omniglot 5-way 1-shot | accuracy: 67.00 -> 83.80 | Helped |
| replace hand-crafted Mini-ImageNet augmentation with Auto Augment (ablation) | UMTRA · Mini-ImageNet 5-way 1-shot | accuracy: 35.09 -> 39.93 | Helped |
| increase translation range from 0 to 0-6 pixels (ablation) | UMTRA with random pixel zeroing · Omniglot 5-way 1-shot | accuracy: 67.0 -> 83.8 | Helped |
| train target tasks from scratch with augmentation rather than UMTRA (ablation) | few-shot classifier · Omniglot 5-way 1-shot | accuracy: 83.8% vs 55.8% | Hurt |
| train target tasks from scratch with augmentation rather than UMTRA (ablation) | few-shot classifier · Mini-ImageNet 5-way 1-shot | accuracy: 39.93% vs 28.8% | Hurt |
Closest papers by meaning
- Unsupervised Learning via Meta-Learning (ICLR 2019 · similarity 0.90)
- Meta-Learning for Semi-Supervised Few-Shot Classification (ICLR 2018 · similarity 0.90)
- Unsupervised Meta-learning via Few-shot Pseudo-supervised Contrastive Learning (ICLR 2023 · similarity 0.89)
- A Universal Representation Transformer Layer for Few-Shot Image Classification (ICLR 2021 · similarity 0.88)
- Unraveling Meta-Learning: Understanding Feature Representations for Few-Shot Tasks (ICML 2020 · similarity 0.88)
- Unsupervised Meta-Learning via In-Context Learning (ICLR 2025 · 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.