Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Learning Graph Embeddings for Open World Compositional Zero-Shot Learning
Massimiliano Mancini, Muhammad Ferjad Naeem, Yongqin Xian, Zeynep Akata
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 |
|---|---|---|---|
| add self-connections for composition nodes in the graph (ablation) | CGE graph · MIT-States val (closed world) | AUC: 7.6 -> 8.1 | Improved |
| add state-object links to the graph (ablation) | CGE graph · MIT-States val (closed world) | AUC: 8.1 -> 8.6 | Improved |
| replace dot-product classifier with cosine classifier (ablation) | CGE · MIT-States val (closed world) | AUC: 8.6 -> 7.9 | No clear effect |
| include all unseen compositions in the loss without margin (ablation) | CompCos · MIT-States val (open world) | AUC: 1.2 -> 1.7 | Improved |
| add feasibility margin alpha>0 (ablation) | CompCos · MIT-States val (open world) | AUC: 1.7 -> 2.0 | Improved |
| linearly warm up alpha (ablation) | CompCos · MIT-States val (open world) | AUC: 2.0 -> 2.1 | Improved |
| average state and object feasibility instead of max (ablation) | CompCos · MIT-States val (open world) | AUC: 2.0 -> 2.1 | Improved |
| keep composition-to-primitive graph weights at 1 instead of feasibility (ablation) | Co-CGE_ff · MIT-States val (open world) | AUC: average improvement of 0.63 in AUC | Improved |
| weight s-o and p-c graph links by feasibility (ablation) | Co-CGE_ff · MIT-States val (open world) | AUC: 2.4 -> 2.5 | Improved |
| fine-tune end-to-end rather than frozen features (ablation) | Co-CGE · MIT-States val (open world) | AUC: 2.5 -> 2.8 | Improved |
| compute feasibility from primitive similarity to training neighbours instead of direct state-object similarity (ablation) | Co-CGE · MIT-States val (open world) | AUC: 1.7 vs 2.5 in AUC | Improved |
| use cosine similarity instead of Euclidean distance for feasibility (ablation) | Co-CGE · MIT-States val (open world) | AUC: 2.5 AUC on Ours vs 1.4 of Exp. Euclidean and 1.1 of Euclidean | Improved |
Closest papers by meaning
- Beyond Seen Primitive Concepts and Attribute-Object Compositional Learning (CVPR 2024 · similarity 0.90)
- A Dynamic Learning Method towards Realistic Compositional Zero-Shot Learning (AAAI 2024 · similarity 0.89)
- Context-based and Diversity-driven Specificity in Compositional Zero-Shot Learning (CVPR 2024 · similarity 0.88)
- Revealing the Proximate Long-Tail Distribution in Compositional Zero-Shot Learning (AAAI 2024 · similarity 0.88)
- ProCC: Progressive Cross-Primitive Compatibility for Open-World Compositional Zero-Shot Learning (AAAI 2024 · similarity 0.88)
- Retrieval-Augmented Primitive Representations for Compositional Zero-Shot Learning (AAAI 2024 · 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.
