Research map / Graph learning / Knowledge and temporal graphs
Learning Graph Embeddings for Compositional Zero-Shot Learning
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 |
|---|---|---|---|
| use the compositional graph with end-to-end feature learning | CGE · MIT-States test | AUC: 3.0 -> 6.5 | Helped |
| use the compositional graph with end-to-end feature learning | CGE · UT-Zappos test | AUC: 29.3 -> 33.5 | Helped |
| fine-tune the image feature extractor rather than keep it fixed (ablation) | CGE · MIT-States test | AUC: 5.1 -> 6.5 | Helped |
| use averaged state and object word embeddings directly instead of learning graph embeddings (ablation) | CGE · MIT-States validation | AUC: 5.9 -> 8.6 | Hurt |
| add self-loops to compositional-label nodes (ablation) | CGE · MIT-States validation | AUC: 7.6 -> 8.1 | Helped |
| add state-object edges to the graph (ablation) | CGE · MIT-States validation | AUC: 8.1 -> 8.6 | Helped |
| add the WordNet object hierarchy to the compositional graph (ablation) | CGE · MIT-States validation | AUC: 8.6 -> 7.9 | Hurt |
| replace the spectral GCN with GCNII (ablation) | CGE · MIT-States validation | AUC: 8.6 -> 7.2 | Hurt |
| increase the GCN depth from 2 to 10 layers (ablation) | CGE_ff · MIT-States validation | AUC: 6.80 -> 4.95 | Hurt |
| increase hidden dimension from 1024 to 4096 with a 2-layer GCN (ablation) | CGE_ff · MIT-States validation | AUC: 6.53 -> 6.80 | Helped |
| increase hidden dimension from 4096 to 8184 with a 2-layer GCN (ablation) | CGE_ff · MIT-States validation | AUC: 6.80 -> 6.59 | Hurt |
| use FastText plus Word2Vec instead of a single embedding model (ablation) | CGE_ff · MIT-States validation | AUC: 6.4 -> 6.8 | Helped |
Closest papers by meaning
- Beyond Seen Primitive Concepts and Attribute-Object Compositional Learning (CVPR 2024 · similarity 0.89)
- A Dynamic Learning Method towards Realistic Compositional Zero-Shot Learning (AAAI 2024 · similarity 0.89)
- Retrieval-Augmented Primitive Representations for Compositional Zero-Shot Learning (AAAI 2024 · similarity 0.89)
- 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)
- Unlocking Compositional Generalization in Continual Few-Shot Learning (NeurIPS 2026 · similarity 0.87)
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.