Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Is Meta-Learning Out? Rethinking Unsupervised Few-Shot Classification with Limited Entropy
Yunchuan Guan, Yu Liu, Ke Zhou, Zhiqi Shen, Jenq-Neng Hwang, Serge Belongie, Lei Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| add DBSCAN heterogeneous task construction instead of K-means (ablation) | MINO · Omniglot 5-way 1-shot | accuracy (%): 87.12 -> 93.81 | Improved |
| add DBSCAN heterogeneous task construction instead of K-means (ablation) | MINO · CIFAR-100 zero-shot | accuracy (%): 37.58 -> 42.34 | Improved |
| use meta-learning instead of whole-class training (ablation) | MINO · Omniglot 5-way 1-shot | accuracy (%): 74.32 -> 93.81 | Improved |
| use meta-learning instead of whole-class training (ablation) | MINO · CIFAR-100 zero-shot | accuracy (%): 32.37 -> 42.34 | Improved |
| add stability-based meta-scaler (ablation) | MINO · Omniglot 5-way 1-shot | accuracy (%): 91.56 -> 93.81 | Improved |
| add stability-based meta-scaler (ablation) | MINO · STL-10 zero-shot | accuracy (%): 56.84 -> 58.74 | Improved |
| use bi-level meta-learning (MAML) instead of single-level WCT under 30% label noise (ablation) | MAML · Omniglot 5-way 1-shot | accuracy (%): 64.65 -> 80.72 | Improved |
| use bi-level meta-learning (MAML) instead of single-level WCT under 30% label noise (ablation) | MAML · Mini-ImageNet 5-way 1-shot | accuracy (%): 29.68 -> 37.51 | Improved |
| add dynamic head for heterogeneous tasks versus static head (ablation) | DHM · Mini-ImageNet 5-10 way heterogeneous | accuracy (%): 41.63 -> 44.09 | Improved |
| use MAML instead of WCT in entropy-limited 3D few-shot (ablation) | MAML · ShapeNetCore 5-way 1-shot | accuracy (%): 49.62 -> 54.03 | Improved |
| use MAML-style instead of ANIL-style inner updates (ablation) | MINO · Omniglot 5-way 1-shot | training time to accuracy: 93.5 reached only by MAML (25717 s); ANIL faster to 90.0 (13987 vs 14723) | Mixed |
| add DBSCAN instead of K-means for pseudo-labels (ablation) | MINO · ImageNet zero-shot | accuracy (%): 29.89 -> 31.12 | Improved |
Closest papers by meaning
- Task Agnostic Meta-Learning for Few-Shot Learning (CVPR 2019 · similarity 0.92)
- Meta-Baseline: Exploring Simple Meta-Learning for Few-Shot Learning (ICCV 2021 · similarity 0.92)
- Unraveling Meta-Learning: Understanding Feature Representations for Few-Shot Tasks (ICML 2020 · similarity 0.91)
- Meta-Learning without Memorization (ICLR 2020 · similarity 0.91)
- Unsupervised Learning via Meta-Learning (ICLR 2019 · similarity 0.91)
- Revisiting Unsupervised Meta-Learning via the Characteristics of Few-Shot Tasks (TPAMI 2022 · similarity 0.91)
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.
