Research map / Label-efficient and robust learning / Semi-supervised learning
Rethinking Clustering-Based Pseudo-Labeling for Unsupervised Meta-Learning
Xingping Dong, Jianbing Shen, Ling Shao
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 |
|---|---|---|---|
| replace ACAI/DC embedding with clustering-friendly embedding (CFE) for k-means pseudo-labels | CACTUs-MAML · Omniglot 5-way 1-shot | accuracy: 68.84 -> 91.32 | Improved |
| replace DC embedding with CFE | CACTUs-MAML · miniImageNet 5-way 1-shot | accuracy: 39.90 -> 41.99 | Improved |
| add negative set without asynchronous embedding (ablation) | CFE-M (positive set only) · miniImageNet 5-way 1-shot | accuracy: 39.75 -> 25.35 | Hurt |
| add asynchronous (momentum) encoder to positive set (ablation) | CFE-M · miniImageNet 5-way 1-shot | accuracy: 39.75 -> 41.04 | Improved |
| add negative set with asynchronous embedding (ablation) | CFE-M (Pos+AE) · miniImageNet 5-way 1-shot | accuracy: 41.04 -> 41.99 | Improved |
| add cluster selection (SC) from progressive evaluation (ablation) | CFE-M (Pos+Neg+AE) · miniImageNet 5-way 1-shot | accuracy: 41.99 -> 43.39 | Improved |
| add noise filtering (FN) on top of cluster selection (ablation) | CFE-M (Pos+Neg+AE+SC) · miniImageNet 5-way 1-shot | accuracy: 43.39 -> 43.60 | Improved |
| add noise filtering without cluster selection (ablation) | CFE-M (Pos+Neg+AE) · miniImageNet 5-way 1-shot | accuracy: 41.99 -> 42.85 | Improved |
| add progressive evaluation (SC+FN) to EP (ablation) | CFE-E (Pos+Neg+AE) · miniImageNet 5-way 1-shot | accuracy: 47.55 -> 49.13 | Improved |
| use CFE (Pos+Neg+AE) instead of MoCo-v2 embedding for pseudo-labels | MAML · miniImageNet 5-way 1-shot | accuracy: 39.83 -> 41.99 | Improved |
| use CFE pseudo-labels with EP instead of CACTUs-EP | EP · Omniglot 20-way 1-shot | accuracy: 53.50 -> 83.67 | Improved |
| use PL-CFE-MAML instead of supervised MAML on many-shot tasks | MAML · miniImageNet 5-way 50-shot | accuracy: 73.77 -> 75.52 | Improved |
Closest papers by meaning
- Semantic-Aware Pseudo-Labeling for Unsupervised Meta-Learning (TPAMI 2025 · similarity 0.93)
- Unsupervised Clustering using Pseudo-semi-supervised Learning (ICLR 2020 · similarity 0.90)
- RPSC: Robust Pseudo-Labeling for Semantic Clustering (AAAI 2024 · similarity 0.87)
- Clustering Properties of Self-Supervised Learning (ICML 2025 · similarity 0.86)
- CGMatch: A Different Perspective of Semi-supervised Learning (CVPR 2025 · similarity 0.86)
- Keep It on a Leash: Controllable Pseudo-label Generation Towards Realistic Long-Tailed Semi-Supervised Learning (NeurIPS 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.
