Research map / Label-efficient and robust learning / Semi-supervised learning
Infinite Mixture Prototypes for Few-shot Learning
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 |
|---|---|---|---|
| add adaptive multi-modal clustering to prototypes (ablation) | Omniglot alphabet recognition · 10-way 10-shot | accuracy: 65.2 $\pm$ 0.6 -> 92.0$\pm$0.1 | Helped |
| learn cluster variance (ablation) | Prototypical networks · mini-ImageNet 5-way 5-shot | accuracy: 66.1 $\pm$ 0.6 -> 67.2 $\pm$ 0.5 | Helped |
| learn separate labeled and unlabeled cluster variances instead of a tied variance (ablation) | IMP · Omniglot semi-supervised 5-way 1-shot | accuracy: 93.5$\pm$0.3 -> 98.9$\pm$0.1 | Helped |
| learn separate labeled and unlabeled cluster variances instead of a tied variance (ablation) | IMP · mini-ImageNet semi-supervised 5-way 1-shot | accuracy: 48.6$\pm$0.4 -> 49.6$\pm$0.8 | Helped |
| use multi-modal prototypes rather than one prototype per class | Omniglot, alphabet training to character testing · 20-way 1-shot | accuracy: 82.1$\pm$0.4 -> 95.4$\pm$0.2 | Helped |
| use IMP rather than prototypes | Omniglot character recognition · 20-way 1-shot, character training and test | accuracy: 94.9$\pm$0.2 -> 95.1$\pm$0.1 | No clear effect |
| use IMP rather than prototypes | Omniglot alphabet recognition · 10-way 10-shot | accuracy: 65.6$\pm$0.4 -> 92.0$\pm$0.1 | No clear effect |
| use two clustering iterations during training rather than one (ablation) | IMP · few-shot classification | accuracy | No clear effect |
| use 100 clustering iterations at inference rather than one (ablation) | IMP · few-shot classification | accuracy | No clear effect |
| replace IMP inference with hard DP-means assignment (ablation) | semi-supervised 5-way 1-shot Omniglot · 5 unlabeled examples per class and 5 distractors | accuracy: 99.0 $\pm$ 0.1 -> 98.0 $\pm$ 0.2 | Hurt |
| replace IMP inference with EM (ablation) | semi-supervised 5-way 1-shot mini-ImageNet · 5 unlabeled examples per class and 5 distractors | accuracy: 49.6 $\pm$ 0.6 -> 41.0 $\pm$ 0.6 | Hurt |
| remove the CRP prior from EM inference (ablation) | EM on mini-ImageNet · semi-supervised 5-way 1-shot | accuracy: 41.0 $\pm$ 0.6 -> 48.6\% | Helped |
Closest papers by meaning
- Prototypical Networks for Few-shot Learning (NeurIPS 2017 · similarity 0.87)
- Self-Organizing Visual Prototypes for Non-Parametric Representation Learning (ICML 2025 · similarity 0.86)
- Provably Improving Generalization of Few-shot models with Synthetic Data (ICML 2025 · similarity 0.85)
- BECLR: Batch Enhanced Contrastive Few-Shot Learning (ICLR 2024 · similarity 0.84)
- Unlocking Compositional Generalization in Continual Few-Shot Learning (NeurIPS 2026 · similarity 0.84)
- Envisioning Class Entity Reasoning by Large Language Models for Few-shot Learning (AAAI 2025 · similarity 0.84)
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.