Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
TAFSSL: Task-Adaptive Feature Sub-Space Learning for few-shot classification
Moshe Lichtenstein, Prasanna Sattigeri, Rogerio Feris, Raja Giryes, Leonid Karlinsky
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 ICA task-adaptive sub-space projection (ablation) | Simple shot NN · miniImageNet 1-shot (transductive) | accuracy (%): 64.30 -> 72.10 | Improved |
| add PCA task-adaptive sub-space projection (ablation) | Simple shot NN · tieredImageNet 1-shot (transductive) | accuracy (%): 71.26 -> 80.07 | Improved |
| use ICA instead of PCA for sub-space (ablation) | TAFSSL · miniImageNet 1-shot (transductive) | accuracy (%): 70.53 -> 72.10 | Improved |
| add Bayesian K-Means after ICA (ablation) | ICA · miniImageNet 1-shot (transductive) | accuracy (%): 72.10 -> 75.79 | Improved |
| add Mean-Shift Propagation after ICA (ablation) | ICA · miniImageNet 1-shot (transductive) | accuracy (%): 72.10 -> 77.06 | Improved |
| add ICA sub-space before clustering (ablation) | MSP · tieredImageNet 1-shot (transductive) | accuracy (%): 76.01 -> 84.29 | Improved |
| add Bayesian K-Means after ICA (ablation) | ICA · miniImageNet 5-shot (transductive) | accuracy (%): 81.85 -> 82.83 | No clear effect |
| add BKM clustering without sub-space projection (ablation) | Simple shot NN · tieredImageNet 5-shot (transductive) | accuracy (%): 86.59 -> 85.67 | Hurt |
| increase unlabeled samples from 30 to 100 (ablation) | ICA + MSP · miniImageNet 1-shot (semi-supervised) | accuracy (%): 78.55 -> 80.11 | Improved |
| use DenseNet backbone instead of Conv-4 (ablation) | PCA + MSP · miniImageNet 1-shot (transductive) | accuracy (%): 56.63 -> 76.31 | Improved |
| increase query set skew up to R = 50 (ablation) | ICA-based TAFSSL · miniImageNet and tieredImageNet transductive | accuracy (%): at most 1% loss at R = 50 | No clear effect |
| add distracting classes to unlabeled data (0 to 7) (ablation) | ICA-based TAFSSL · semi-supervised miniImageNet | accuracy (%): about 8% gap kept | No clear effect |
Closest papers by meaning
- Few-Shot Learning With a Strong Teacher (TPAMI 2022 · similarity 0.91)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.90)
- Pushing the Limits of Simple Pipelines for Few-Shot Learning: External Data and Fine-Tuning Make a Difference (CVPR 2022 · similarity 0.90)
- Few-Shot Learning With Noisy Labels (CVPR 2022 · similarity 0.90)
- Discriminative Sample-Guided and Parameter-Efficient Feature Space Adaptation for Cross-Domain Few-Shot Learning (CVPR 2024 · similarity 0.89)
- Scaling Few-Shot Learning for the Open World (AAAI 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.
