Research map / Label-efficient and robust learning / Recognition and long-tail
Learning to Learn Image Classifiers With Visual Analogy
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 |
|---|---|---|---|
| combine transferred and data-learned classifier parameters by voting (ablation) | VAGER · ImageNet novel classes, binary classification | AUC: 0.8556 -> 0.8718 | Helped |
| use voting rather than transferred initialization (ablation) | VAGER · ImageNet novel classes, 1-shot binary | AUC: 0.7662 -> 0.8718 | Helped |
| use voting rather than parameter tuning toward transferred parameters (ablation) | VAGER · ImageNet novel classes, 1-shot binary | AUC: 0.7923 -> 0.8718 | Helped |
| add the embedding-to-classifier mapping regression term (ablation) | VAGER · ImageNet novel classes, 1-shot binary | AUC: 0.8261 -> 0.8718 | Helped |
| add visual analogy graph embedding (ablation) | VAGER · ImageNet novel classes, 1-shot binary | AUC: 0.7922 -> 0.8718 | Helped |
| transfer classifier parameters from related base classes | VAGER · ImageNet novel classes, 1-shot binary | AUC: 0.7705 -> 0.8718 | Helped |
| generalize from base classes to the novel Bubble class | VAGER · Bubble, 1-shot binary | AUC: 0.7455 -> 0.7041 | Hurt |
| increase the number of training shots for Weighted-LR | Weighted-LR · ImageNet novel classes, binary classification | AUC: 0.8440; 0.8458; 0.8509; 0.8468 | No clear effect |
| use VAGER+Voting instead of the strongest listed baseline in each group | ImageNet few-shot classifier · ImageNet novel classes, 10/30/50/100-way 1-shot | top-1 accuracy: 67.59% vs 65.69%; 63.96% vs 62.42%; 58.02% vs 57.13%; 51.27% vs 48.85%; 56.24% vs 54.54%; 40.73% vs 40.15%; 38.69% vs 37 | Helped |
| use classifier transfer from base classes rather than LR alone | VAGER · ImageNet novel classes, 1-shot binary | AUC: Jeep 0.8034 -> 0.9469; Bubble 0.7455 -> 0.7041 | Mixed |
Closest papers by meaning
- Online Collaborative Learning for Open-Vocabulary Visual Classifiers (CVPR 2016 · similarity 0.87)
- L-WISE: Boosting Human Visual Category Learning Through Model-Based Image Selection and Enhancement (ICLR 2025 · similarity 0.85)
- Language-Informed Visual Concept Learning (ICLR 2024 · similarity 0.84)
- Language Model as Visual Explainer (NeurIPS 2024 · similarity 0.84)
- $\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs (ICLR 2025 · similarity 0.84)
- ConceptBed: Evaluating Concept Learning Abilities of Text-to-Image Diffusion Models (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.