Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Preserving Semantic Relations for Zero-Shot Learning
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 |
|---|---|---|---|
| preserve category-level semantic relations in the embedding space | zero-shot recognition · SUN, AWA2, CUB, aPY conventional ZSL | average per-class top-1 accuracy (%): 61.4 vs 58.1; 63.8 vs 62.5; 56.0 vs 54.9; 38.4 vs 39.8 | Mixed |
| preserve category-level semantic relations in the embedding space | zero-shot recognition · SUN, AWA2, CUB, aPY generalized ZSL | H: 26.7 vs 26.3; 32.3 vs 27.8; 33.9 vs 34.4; 21.4 vs 19.0 | Mixed |
| use relation-based objectives instead of mean squared error alignment (ablation) | Proposed zero-shot model · SUN conventional ZSL | average per-class top-1 accuracy (%): 58.5 -> 61.4 | Helped |
| use relation-based objectives instead of mean squared error alignment (ablation) | Proposed zero-shot model · AWA2 conventional ZSL | average per-class top-1 accuracy (%): 54.9 -> 63.8 | Helped |
| use relation-based objectives instead of mean squared error alignment (ablation) | Proposed zero-shot model · CUB conventional ZSL | average per-class top-1 accuracy (%): 49.2 -> 56.0 | Helped |
| use relation-based objectives instead of mean squared error alignment (ablation) | Proposed zero-shot model · aPY conventional ZSL | average per-class top-1 accuracy (%): 34.8 -> 38.4 | Helped |
| add the similar-class relation objective O2 (ablation) | Proposed zero-shot model · SUN conventional ZSL | average per-class top-1 accuracy (%): 57.1 -> 61.4 | Helped |
| add the similar-class relation objective O2 (ablation) | Proposed zero-shot model · AWA2 conventional ZSL | average per-class top-1 accuracy (%): 57.2 -> 63.8 | Helped |
| add the similar-class relation objective O2 (ablation) | Proposed zero-shot model · CUB conventional ZSL | average per-class top-1 accuracy (%): 51.5 -> 56.0 | Helped |
| add the similar-class relation objective O2 (ablation) | Proposed zero-shot model · aPY conventional ZSL | average per-class top-1 accuracy (%): 31.6 -> 38.4 | Helped |
| add the reconstruction objective O3 (ablation) | Proposed zero-shot model · SUN conventional ZSL | average per-class top-1 accuracy (%): 58.7 -> 61.4 | Helped |
| add the reconstruction objective O3 (ablation) | Proposed zero-shot model · AWA2 conventional ZSL | average per-class top-1 accuracy (%): 62.4 -> 63.8 | Helped |
Closest papers by meaning
- Learning a Deep Embedding Model for Zero-Shot Learning (CVPR 2017 · similarity 0.91)
- VGSE: Visually-Grounded Semantic Embeddings for Zero-Shot Learning (CVPR 2022 · similarity 0.91)
- Deep Semantic Structural Constraints for Zero-Shot Learning (AAAI 2018 · similarity 0.91)
- Generalized Zero-Shot Recognition Based on Visually Semantic Embedding (CVPR 2019 · similarity 0.91)
- TPR: Topology-Preserving Reservoirs for Generalized Zero-Shot Learning (NeurIPS 2024 · similarity 0.89)
- ZeroDiff: Solidified Visual-semantic Correlation in Zero-Shot Learning (ICLR 2025 · similarity 0.89)
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.