Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Shallow Bayesian Meta Learning for Real-World Few-Shot Recognition
Xueting Zhang, Debin Meng, Henry Gouk, Timothy M. Hospedales
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 a hand-crafted NIW prior to QDA (ablation) | QDA · miniImageNet 5-shot (Conv-4) | accuracy: 34.45 -> 71.40 | Improved |
| meta-learn the NIW prior instead of hand-crafting it (ablation) | QDA with prior · miniImageNet 5-shot (Conv-4) | accuracy: 71.40 -> 72.64 | Improved |
| meta-learn the NIW prior instead of hand-crafting it (ablation) | QDA with prior · miniImageNet 5-shot (WRN) | accuracy: 81.18 -> 84.28 | Improved |
| use QDA instead of LDA with meta-learned prior (ablation) | MetaQDA · miniImageNet 1-shot (WRN) | accuracy: 64.92 -> 67.83 | Improved |
| use QDA instead of LDA with meta-learned prior (ablation) | MetaQDA · miniImageNet 5-shot (Conv-4) | accuracy: 72.39 -> 72.64 | No clear effect |
| use Bayesian meta-learning instead of MAML (ablation) | MetaQDA · miniImageNet 1-shot (Conv-4) | accuracy: 54.33 -> 56.41 | Improved |
| use Bayesian meta-learning instead of MAML (ablation) | MetaQDA · miniImageNet 5-shot (ResNet-18) | accuracy: 77.82 -> 80.98 | Improved |
| use Bayesian meta-learning instead of MAML (ablation) | MetaQDA · miniImageNet 1-shot (Conv-4) | ECE: 52.75 -> 8.03 | Improved |
| use fully Bayesian instead of MAP inference (ablation) | MetaQDA · miniImageNet 1-shot (Conv-4) | ECE+TS: 2.75 -> 2.33 | Improved |
| use fully Bayesian instead of MAP inference (ablation) | MetaQDA · miniImageNet 5-shot (WRN) | ECE: 17.37 -> 15.86 | Improved |
| add MetaQDA classifier to URT features | URT+PN · Meta-Dataset | average accuracy overall: 73.7 -> 74.3 | Improved |
| replace SimpleShot NCC with MetaQDA in class-incremental learning | NCC · miniImageNet FSCIL session 8 | accuracy: 32.26 -> 35.78 | Improved |
Closest papers by meaning
- Meta-Baseline: Exploring Simple Meta-Learning for Few-Shot Learning (ICCV 2021 · similarity 0.90)
- Meta-Learning With Differentiable Convex Optimization (CVPR 2019 · similarity 0.90)
- Meta-Learning to Detect Rare Objects (ICCV 2019 · similarity 0.90)
- Unraveling Meta-Learning: Understanding Feature Representations for Few-Shot Tasks (ICML 2020 · similarity 0.90)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.88)
- MetaCoCo: A New Few-Shot Classification Benchmark with Spurious Correlation (ICLR 2024 · similarity 0.88)
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.
