Research map / Label-efficient and robust learning / Semi-supervised learning
PLATINUM: Semi-Supervised Model Agnostic Meta-Learning using Submodular Mutual Information
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 |
|---|---|---|---|
| SMI-based per-class semi-supervision (FLMI/GCMI) added to MAML inner and outer loops | MAML · mini-ImageNet 1-shot w/o OOD, rho=1% | accuracy (%): +7.01 (FLMI 42.27 vs MAML 35.26) | Helped |
| SMI-based per-class semi-supervision added to MAML | MAML · mini-ImageNet 5-shot w/o OOD, rho=1% | accuracy (%): +3.58 (FLMI 63.80 vs MAML 60.22) | Helped |
| Pseudo-labeling added to MAML inner and outer loops | MAML · mini-ImageNet 1-shot w/o OOD, rho=1% | accuracy (%): +2.45 (PL 37.71 vs MAML 35.26) | Helped |
| SMI-based semi-supervision added to MAML under OOD classes in unlabeled set | MAML · mini-ImageNet 1-shot w/ OOD, rho=1% | accuracy (%): +7.31 (GCMI 42.57 vs MAML 35.26) | Helped |
| Pseudo-labeling added to MAML under OOD classes in unlabeled set | MAML · mini-ImageNet 1-shot w/ OOD, rho=1% | accuracy (%): -0.10 relative to MAML (PL 35.16 vs 35.26) | Helped |
| Pseudo-labeling added to MAML under OOD classes at 5-shot | MAML · mini-ImageNet 5-shot w/ OOD, rho=1% | accuracy (%): PL 60.31 vs MAML 60.20 | No clear effect |
| Adding SMI semi-supervision on tieredImageNet 1-shot | MAML · tiered-ImageNet 1-shot w/o OOD, rho=1% | accuracy (%): +3.67 (FLMI 45.63 vs MAML 41.96) | Helped |
| Outer-loop SMI/pseudo-label selection vs none (ablation) | PL, GCMI, FLMI · mini-ImageNet 1-shot and 5-shot | accuracy (%): ~4% gain for PL 1-shot (text only) | Helped |
| Increasing number of OOD distractor classes (SMI methods vs PL) (ablation) | Flmi, Gcmi vs PL · mini-ImageNet 5-shot, varying OOD classes | accuracy (%): PL falls to slightly below MAML at many OOD classes; SMI methods stay above MAML | Mixed |
| Gcmi vs Flmi under OOD classes on tieredImageNet/CIFAR-FS (ablation) | SMI variants · OOD-containing unlabeled sets | accuracy (%): Gcmi slightly better on some OOD 5-shot settings (62.59 vs 62.19 tiered; 54.42 vs 54.16 CIFAR-FS) | Mixed |
| Larger labeled ratio rho (lower gain from SSL) on mini-ImageNet (ablation) | PLATINUM vs baselines · mini-ImageNet 1-shot, rho 1% to 30% | accuracy (%): gains shrink as rho grows; at 30% Flmi 48.98 vs PL 44.37, MAML 45.26 (gain still present) | Mixed |
Closest papers by meaning
- Learning to Self-Train for Semi-Supervised Few-Shot Classification (NeurIPS 2019 · similarity 0.90)
- Meta-Learning for Semi-Supervised Few-Shot Classification (ICLR 2018 · similarity 0.89)
- Unsupervised Meta-learning via Few-shot Pseudo-supervised Contrastive Learning (ICLR 2023 · similarity 0.88)
- Task Cooperation for Semi-Supervised Few-Shot Learning (AAAI 2021 · similarity 0.88)
- MetaCoCo: A New Few-Shot Classification Benchmark with Spurious Correlation (ICLR 2024 · similarity 0.87)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · 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.