Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Joint Distribution Matters: Deep Brownian Distance Covariance for Few-Shot Classification
Check 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 |
|---|---|---|---|
| Use BDC representations instead of marginal-moment representations | DeepBDC versus ProtoNet, CovNet and ADM · miniImageNet 5-way 1-shot and 5-shot | accuracy: 67.34$\pm$0.43 vs 62.11$\pm$0.44; 84.46$\pm$0.28 vs 82.05$\pm$0.29 | Helped |
| Use STL DeepBDC rather than Good-Embed | STL DeepBDC · miniImageNet 5-way 1-shot and 5-shot | accuracy: 67.83$\pm$0.43 vs 64.82$\pm$0.44; 85.45$\pm$0.29 vs 82.14$\pm$0.43 | Helped |
| Use BDC joint-distribution representations | STL DeepBDC · tieredImageNet and CUB | accuracy | Helped |
| Set Meta DeepBDC channel dimension to 640 (ablation) | Meta DeepBDC · miniImageNet 5-way 1-shot and 5-shot | accuracy: 67.34$\pm$0.43; 84.46$\pm$0.28 | Helped |
| Replace inner product with cosine similarity (ablation) | Meta DeepBDC · miniImageNet 5-way 1-shot | accuracy: 67.34$\pm$0.43 -> 61.74$\pm$0.42 | Hurt |
| Replace inner product with Euclidean distance (ablation) | Meta DeepBDC · miniImageNet 5-way 1-shot | accuracy: 67.34$\pm$0.43 -> 56.70$\pm$0.45 | Hurt |
| Replace inner product with Euclidean distance (ablation) | Meta DeepBDC · miniImageNet 5-way 5-shot | accuracy: 84.46$\pm$0.28 -> 84.46$\pm$0.28 | No clear effect |
| Set STL DeepBDC channel dimension to 128 (ablation) | STL DeepBDC · miniImageNet 5-way 1-shot and 5-shot | accuracy: 67.83$\pm$0.43; 85.45$\pm$0.30 | Helped |
| Replace logistic regression with SVM (ablation) | STL DeepBDC · miniImageNet 5-way 1-shot and 5-shot | accuracy: 67.83$\pm$0.43 -> 66.29$\pm$0.44; 85.45$\pm$0.30 -> 84.73$\pm$0.29 | Hurt |
| Replace logistic regression with softmax classifier (ablation) | STL DeepBDC · miniImageNet 5-way 1-shot and 5-shot | accuracy: 67.83$\pm$0.43 -> 66.30$\pm$0.44; 85.45$\pm$0.30 -> 85.20$\pm$0.29 | Mixed |
| Average support features before computing the prototype BDC matrix (ablation) | Meta DeepBDC · miniImageNet 5-way 5-shot | accuracy: 84.46 -> 82.36 | Hurt |
| Concatenate support features before computing the prototype BDC matrix (ablation) | Meta DeepBDC · miniImageNet 5-way 5-shot | accuracy: 84.46 -> 83.74 | Hurt |
Closest papers by meaning
- Improved Few-Shot Visual Classification (CVPR 2020 · similarity 0.89)
- Distribution Consistency Based Covariance Metric Networks for Few-Shot Learning (AAAI 2019 · similarity 0.89)
- Revisiting Local Descriptor Based Image-To-Class Measure for Few-Shot Learning (CVPR 2019 · similarity 0.89)
- DMN4: Few-Shot Learning via Discriminative Mutual Nearest Neighbor Neural Network (AAAI 2022 · similarity 0.88)
- Learning To Affiliate: Mutual Centralized Learning for Few-Shot Classification (CVPR 2022 · similarity 0.88)
- A Baseline for Few-Shot Image Classification (ICLR 2020 · 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.