Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
TADAM: Task dependent adaptive metric for improved few-shot learning
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 |
|---|---|---|---|
| add learnable scaling to cosine similarity (ablation) | Prototypical Networks · mini-Imagenet 5-way train, 5-shot | accuracy: 54.5 $\pm$ 1.1 -> 68.2 $\pm$ 0.8 | Helped |
| add learnable scaling to cosine similarity (ablation) | Prototypical Networks · FC100 5-way train, 5-shot | accuracy: 40.9 $\pm$ 0.6 -> 51.0 $\pm$ 0.6 | Helped |
| add metric scaling (ablation) | ResNet-12 few-shot model · mini-Imagenet 5-way 5-shot | accuracy: 74.2 $\pm$ 0.2 -> 75.7 $\pm$ 0.2 | Helped |
| add metric scaling (ablation) | ResNet-12 few-shot model · FC100 5-way 5-shot | accuracy: 53.3 $\pm$ 0.5 -> 54.0 $\pm$ 0.5 | Helped |
| add auxiliary classification co-training (ablation) | scaled Euclidean ResNet-12 · mini-Imagenet 5-way 5-shot | accuracy: 75.7 $\pm$ 0.2 -> 75.6 $\pm$ 0.4 | No clear effect |
| add auxiliary classification co-training (ablation) | scaled Euclidean ResNet-12 · FC100 5-way 5-shot | accuracy: 54.0 $\pm$ 0.5 -> 54.7 $\pm$ 0.5 | Helped |
| add TEN task conditioning without auxiliary co-training (ablation) | scaled Euclidean ResNet-12 · mini-Imagenet 5-way 5-shot | accuracy: 75.7 $\pm$ 0.2 -> 74.6 $\pm$ 0.3 | Hurt |
| add TEN task conditioning without auxiliary co-training (ablation) | scaled Euclidean ResNet-12 · FC100 5-way 5-shot | accuracy: 54.0 $\pm$ 0.5 -> 54.0 $\pm$ 0.7 | No clear effect |
| combine auxiliary classification co-training and TEN task conditioning (ablation) | scaled Euclidean ResNet-12 · mini-Imagenet 5-way 5-shot | accuracy: 75.7 $\pm$ 0.2 -> 76.7 $\pm$ 0.3 | Helped |
| combine auxiliary classification co-training and TEN task conditioning (ablation) | scaled Euclidean ResNet-12 · FC100 5-way 5-shot | accuracy: 54.0 $\pm$ 0.5 -> 56.1 $\pm$ 0.4 | Helped |
| add TEN conditioning in layers preceding max-pooling (ablation) | TADAM feature extractor · mini-Imagenet | accuracy | Helped |
| add TEN conditioning only in the very last layer (ablation) | TADAM feature extractor · mini-Imagenet | accuracy | Helped |
Closest papers by meaning
- Finding Task-Relevant Features for Few-Shot Learning by Category Traversal (CVPR 2019 · similarity 0.90)
- Boosting Few-Shot Learning With Adaptive Margin Loss (CVPR 2020 · similarity 0.89)
- Adaptive Poincaré Point to Set Distance for Few-Shot Classification (AAAI 2022 · similarity 0.89)
- Looking Wider for Better Adaptive Representation in Few-Shot Learning (AAAI 2021 · similarity 0.89)
- A Baseline for Few-Shot Image Classification (ICLR 2020 · similarity 0.88)
- Dense Classification and Implanting for Few-Shot Learning (CVPR 2019 · 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.