Research map / Label-efficient and robust learning / Semi-supervised learning
Iterative Label Cleaning for Transductive and Semi-Supervised Few-Shot Learning
Michalis Lazarou, Tania Stathaki, Yannis Avrithis
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 label propagation using unlabeled queries (transductive vs inductive) (ablation) | iLPC · miniImageNet 1-shot transductive (WRN-28-10) | accuracy (%): 68.17 -> 74.24 | Improved |
| add class balancing to label propagation (ablation) | iLPC · miniImageNet 1-shot transductive (WRN-28-10) | accuracy (%): 74.24 -> 79.42 | Improved |
| add iterative label cleaning to label propagation and balancing (ablation) | iLPC · miniImageNet 1-shot transductive (WRN-28-10) | accuracy (%): 79.42 -> 83.05 | Improved |
| add iterative label cleaning without balancing (ablation) | LP · miniImageNet 1-shot transductive (WRN-28-10) | accuracy (%): 74.24 -> 78.29 | Improved |
| replace iterative label cleaning with selecting column-wise maxima of P (iProb) (ablation) | iLPC · miniImageNet 1-shot transductive (WRN-28-10) | accuracy (%): 83.05 -> 80.75 | Hurt |
| replace label propagation with a linear classifier (ablation) | iLPC · miniImageNet 1-shot transductive (WRN-28-10) | accuracy (%): 83.05 -> 82.04 | Hurt |
| use true class prior for balancing instead of none in unbalanced queries (ablation) | iLPC · miniImageNet 1-shot unbalanced transductive | accuracy (%): 78.06 -> 82.68 | Improved |
| use uniform balancing when queries are unbalanced (ablation) | iLPC · miniImageNet 1-shot unbalanced transductive | accuracy (%): 78.06 -> 77.50 | Hurt |
| add PT+MAP feature pre-processing (power transform, normalization, centering) (ablation) | PT+MAP · miniImageNet 1-shot transductive | accuracy (%): 48.57 -> 82.88 | Improved |
| add PT+MAP feature pre-processing (power transform, normalization, centering) (ablation) | iLPC · miniImageNet 1-shot transductive | accuracy (%): 78.89 -> 83.05 | Improved |
| increase queries from 15 to 50 per class | iLPC vs PT+MAP · miniImageNet 1-shot transductive | accuracy (%): 83.05 -> 85.98 | Improved |
| use ResNet-12B with no augmentation or scaling, iLPC vs MCT | iLPC vs MCT · miniImageNet 1-shot transductive | accuracy (%): 75.58 vs 67.26 | Improved |
Closest papers by meaning
- An Embarrassingly Simple Approach to Semi-Supervised Few-Shot Learning (NeurIPS 2022 · similarity 0.91)
- Negatives Make a Positive: An Embarrassingly Simple Approach to Semi-Supervised Few-Shot Learning (TPAMI 2023 · similarity 0.91)
- Barely-Supervised Learning: Semi-supervised Learning with Very Few Labeled Images (AAAI 2022 · similarity 0.90)
- Learning with Partial-Label and Unlabeled Data: A Uniform Treatment for Supervision Redundancy and Insufficiency (ICML 2024 · similarity 0.88)
- Label-Efficient Dataset Pruning via Semi-Supervised Pseudo-Labeling (NeurIPS 2026 · similarity 0.87)
- Keep It on a Leash: Controllable Pseudo-label Generation Towards Realistic Long-Tailed Semi-Supervised Learning (NeurIPS 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.
