Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Mixture-Based Feature Space Learning for Few-Shot Image Classification
Arman Afrasiyabi, Jean-François Lalonde, Christian Gagné
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 assignment loss La only to the baseline (ablation) | MixtFSL (ResNet-12) · miniImageNet validation 1-shot | accuracy: 56.55 -> 56.52 | No clear effect |
| add diversity loss Ld with stop-gradient (initial training) (ablation) | MixtFSL (ResNet-12) · miniImageNet validation 1-shot | accuracy: 56.52 -> 57.88 | Improved |
| add diversity loss Ld with stop-gradient (initial training) (ablation) | MixtFSL (ResNet-18) · miniImageNet validation 1-/5-shot | accuracy: 72.67 -> 69.43 | Mixed |
| add progressive following stage with Lpf (ablation) | MixtFSL (ResNet-12) · miniImageNet validation 5-shot | accuracy: 73.94 -> 76.09 | Improved |
| remove stop-gradient from diversity loss (ablation) | MixtFSL · miniImageNet base classes | component collapse | Hurt |
| increase initial components per class from 5 to 10 (ablation) | MixtFSL (ResNet-12) · miniImageNet 1-shot | accuracy: 62.29 -> 64.01 | Improved |
| increase initial components per class from 15 to 20 (ablation) | MixtFSL (ResNet-12) · miniImageNet 1-shot | accuracy: 63.98 -> 63.91 | No clear effect |
| use negative margin instead of positive margin (ablation) | MixtFSL (ResNet-12) · miniImageNet 1-shot | accuracy: 63.57 -> 63.98 | No clear effect |
| use margin m=-0.02 instead of m=+0.01 (ablation) | MixtFSL (ResNet-12) · miniImageNet validation 5-shot | accuracy: 77.43 -> 80.38 | Improved |
| use temperature tau=0.05 instead of tau=1 (ablation) | MixtFSL (ResNet-12, ResNet-18) · miniImageNet validation | accuracy | Improved |
| replace centroid alignment features with MixtFSL features | Centroid Alignment (ResNet-12) · miniImageNet 5-shot | accuracy: 80.96 -> 82.45 | Improved |
| replace centroid alignment features with MixtFSL features | Centroid Alignment (ResNet-18) · tieredImageNet 5-shot | accuracy: 85.97 -> 85.57 | Hurt |
Closest papers by meaning
- Discriminative Sample-Guided and Parameter-Efficient Feature Space Adaptation for Cross-Domain Few-Shot Learning (CVPR 2024 · similarity 0.88)
- Unsupervised Few-Shot Image Classification by Learning Features into Clustering Space (ECCV 2022 · similarity 0.88)
- TAFSSL: Task-Adaptive Feature Sub-Space Learning for few-shot classification (ECCV 2020 · 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)
- Provably Learning Diverse Features in Multi-View Data with Midpoint Mixup (ICML 2023 · similarity 0.86)
- Few-shot Learning via Dirichlet Tessellation Ensemble (ICLR 2022 · 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.
