Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Glocal Energy-Based Learning for Few-Shot Open-Set Recognition
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 class-wise and pixel-wise glocal energy scoring (ablation) | GEL · miniImageNet 5-way 1-shot | AUROC: 69.85 ± 0.87 -> 73.70 ± 0.82 | Helped |
| Add class-wise and pixel-wise glocal energy scoring (ablation) | GEL · tieredImageNet 5-way 1-shot | AUROC: 71.66 ± 0.82 -> 75.86 ± 0.81 | Helped |
| Add margin-based energy loss (ablation) | GEL baseline · miniImageNet 5-way 1-shot | AUROC: 69.85 ± 0.87 -> 71.97 ± 0.83 | Helped |
| Add margin-based energy loss (ablation) | GEL baseline · tieredImageNet 5-way 1-shot | AUROC: 71.66 ± 0.82 -> 73.57 ± 0.84 | Helped |
| Add pixel-wise similarity (ablation) | GEL baseline · miniImageNet 5-way 1-shot | AUROC: 69.85 ± 0.87 -> 72.49 ± 0.76 | Helped |
| Add pixel-wise similarity (ablation) | GEL baseline · tieredImageNet 5-way 1-shot | AUROC: 71.66 ± 0.82 -> 74.06 ± 0.73 | Helped |
| Combine class-wise and pixel-wise similarities before energy calculation rather than delaying combination (ablation) | GEL · miniImageNet 5-way 1-shot | AUROC: 73.17 ± 0.83 -> 73.70 ± 0.82 | Helped |
| Combine class-wise and pixel-wise similarities before energy calculation rather than delaying combination (ablation) | GEL · miniImageNet 5-way 5-shot | AUROC: 80.67 ± 0.65 -> 82.29 ± 0.60 | Helped |
| Use learnable rather than fixed equal energy-combination coefficients (ablation) | GEL · miniImageNet 5-way 5-shot | AUROC: 82.29 ± 0.60 -> 81.37 ± 0.64 | Hurt |
| Use task-adaptive rather than fixed equal energy-combination coefficients (ablation) | GEL · miniImageNet 5-way 5-shot | AUROC: 82.29 ± 0.60 -> 81.34 ± 0.62 | Hurt |
| Use learnable rather than fixed equal energy-combination coefficients (ablation) | GEL · tieredImageNet 5-way 5-shot | AUROC: 81.95 ± 0.72 -> 80.86 ± 0.70 | Hurt |
| Use task-adaptive rather than fixed equal energy-combination coefficients (ablation) | GEL · tieredImageNet 5-way 5-shot | AUROC: 81.95 ± 0.72 -> 80.49 ± 0.72 | Hurt |
Closest papers by meaning
- Task-Adaptive Negative Envision for Few-Shot Open-Set Recognition (CVPR 2022 · similarity 0.89)
- Few-Shot Open-Set Recognition Using Meta-Learning (CVPR 2020 · similarity 0.86)
- Few-Shot Open-Set Recognition by Transformation Consistency (CVPR 2021 · similarity 0.86)
- Meta Evidential Transformer for Few-Shot Open-Set Recognition (ICML 2024 · similarity 0.85)
- Frequency Shuffling and Enhancement for Open Set Recognition (AAAI 2024 · similarity 0.84)
- All Beings Are Equal in Open Set Recognition (AAAI 2024 · similarity 0.84)
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.