Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Exploring Active Learning in Meta-Learning: Enhancing Context Set Labeling
Wonho Bae, Jing Wang, Danica J. Sutherland
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 |
|---|---|---|---|
| select context set with Gaussian mixture centers instead of random points | ProtoNet · FC100 5-way 1-shot (unstratified) | accuracy (%): 31.40 -> 35.03 | Improved |
| use uncertainty-based selection (Entropy or Margin) instead of random | MAML · MiniImageNet 5-way 5-shot (unstratified) | accuracy (%): 58.84 -> 39.57 (Entropy) | Hurt |
| use GMM selection instead of random for Baseline++ | Baseline++ · CUB 5-way 5-shot (test unstratified) | accuracy (%): 79.57 -> 82.68 | Improved |
| use active selection at meta-train time (ablation) | meta-learners · meta-training context selection | final predictor quality | Hurt |
| use GMM instead of hybrid uncertainty+diversity methods | MAML · MiniImageNet 5-way 1-shot (train strat.) | accuracy (%): 30.59 -> 33.34 vs k-means Entropy | Improved |
| use GMM instead of k-means for selection (ablation) | Baseline++ · CUB 5-way | accuracy (points): +3.95 (1-shot), +11.79 (5-shot) | Improved |
| use meta-learned features instead of SimCLR features for selection (ablation) | GMM selection · few-shot classification | accuracy | Improved |
| use GMM selection in regression instead of random | MAML · Sine regression 3-shot | MSE: 24.17 -> 18.09 | Improved |
| use GMM selection in cross-domain few-shot | ImageNet-pretrained ResNet18 · Places 1-shot | accuracy (%): 44.28 -> 60.01 | Improved |
| use GMM instead of ProbCover on cross-domain CUB 1-shot | ImageNet-pretrained ResNet18 · Cross-domain CUB 1-shot | accuracy (%): 62.13 -> 59.87 | Hurt |
Closest papers by meaning
- Advances and Challenges in Meta-Learning: A Technical Review (TPAMI 2024 · similarity 0.88)
- Contextual Active Model Selection (NeurIPS 2024 · similarity 0.86)
- Continuous Meta-Learning without Tasks (NeurIPS 2020 · similarity 0.86)
- Robust Meta-Representation Learning via Global Label Inference and Classification (TPAMI 2023 · similarity 0.86)
- Context-Aware Meta-Learning (ICLR 2024 · similarity 0.86)
- Composite Active Learning: Towards Multi-Domain Active Learning with Theoretical Guarantees (AAAI 2024 · similarity 0.85)
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.
