Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
SEN: A Novel Feature Normalization Dissimilarity Measure for Prototypical Few-Shot Learning Networks
Van Nhan Nguyen, Sigurd Løkse, Kristoffer Wickstrøm, Michael Kampffmeyer, Davide Roverso, Robert Jenssen
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 |
|---|---|---|---|
| replace squared Euclidean distance with SEN dissimilarity | Prototypical Network (4CONV) · Mini-ImageNet 5-way 5-shot | accuracy: 67.8% -> 69.8% | Improved |
| replace squared Euclidean distance with SEN dissimilarity (ablation) | Prototypical Network (4CONV) · FC100 5-way 5-shot | accuracy: 52.4% -> 54.6% | Improved |
| replace squared Euclidean distance with SEN dissimilarity (ablation) | Prototypical Network (4CONV) · Omniglot 20-way 5-shot | accuracy: 98.6% -> 98.8% | Improved |
| replace squared Euclidean distance with SEN dissimilarity | Prototypical Network (WRN) · Mini-ImageNet | accuracy: 71.0% -> 72.3% | Improved |
| replace squared Euclidean distance with SEN dissimilarity | Prototypical Network (WRN) · Omniglot | accuracy: 99.2% -> 99.4% | Improved |
| use SEN instead of Ring loss (ablation) | Prototypical Network · Mini-ImageNet 5-way 5-shot | accuracy: 68.6% -> 69.8% | Improved |
| use SEN instead of Ring loss (ablation) | Prototypical Network · FC100 5-way 5-shot | accuracy: 52.8% -> 54.6% | Improved |
| add Ring loss | Prototypical Network · Mini-ImageNet 5-way 5-shot | accuracy: 67.8% -> 68.6% | Improved |
| use higher Ring loss weight gamma (ablation) | Ring PN · Mini-ImageNet | accuracy | Hurt |
| test SEN-trained model with Euclidean distance instead of SEN (ablation) | SEN PN · Mini-ImageNet 5-way 5-shot | accuracy: 69.8% -> 68.8% | Hurt |
| test SEN-trained model with cosine distance instead of SEN (ablation) | SEN PN · Mini-ImageNet 5-way 5-shot | accuracy: 69.8% -> 69.8% | No clear effect |
| train with cosine distance instead of Euclidean (ablation) | Prototypical Network · Mini-ImageNet 5-way 5-shot | accuracy: 67.8% -> 53.3% | Hurt |
Closest papers by meaning
- Prototypical Networks for Few-shot Learning (NeurIPS 2017 · similarity 0.90)
- Revisiting Prototypical Network for Cross Domain Few-Shot Learning (CVPR 2023 · similarity 0.88)
- Kernel Relative-Prototype Spectral Filtering for Few-Shot Learning (ECCV 2022 · similarity 0.87)
- DMN4: Few-Shot Learning via Discriminative Mutual Nearest Neighbor Neural Network (AAAI 2022 · similarity 0.87)
- DiffKendall: A Novel Approach for Few-Shot Learning with Differentiable Kendall's Rank Correlation (NeurIPS 2023 · similarity 0.86)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.85)
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.
