Research map / Large language models / Language, safety and interpretability
Efficient and Effective In-context Demonstration Selection with Coreset
Zihua Wang, Jiarui Wang, Haiyang Xu, Ming Yan, Fei Huang, Xu Yang, Xiu-Shen Wei, Siya Mi, Yu Zhang
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 |
|---|---|---|---|
| use cluster-pruning coreset instead of random coreset (ablation) | CoDR · MSCOCO captioning 4-shot (OFv2) | CIDEr: 80.12 -> 104.23 | Improved |
| use cluster-pruning coreset instead of centroid coreset (ablation) | CoDR · MSCOCO captioning 4-shot (OFv2) | CIDEr: 101.09 -> 104.23 | Improved |
| use cluster-pruning coreset instead of centroid coreset (ablation) | CoDR · Stanford Dogs FIC 4-shot (IDEv2) | accuracy: 63.56 -> 66.28 | Improved |
| use dual retrieval over full support set instead of random pick in coreset (ablation) | CoDR · VQAv2 4-shot (OFv2) | accuracy: 43.58 -> 53.49 | Improved |
| use dual retrieval instead of similarity pick in coreset (ablation) | CoDR · Stanford Dogs FIC 4-shot (OFv2) | accuracy: 36.47 -> 41.39 | Improved |
| use dual retrieval instead of diversity-only pick in coreset (ablation) | CoDR · Stanford Dogs FIC 4-shot (OFv2) | accuracy: 24.44 -> 41.39 | Improved |
| use dual retrieval instead of similarity pick in coreset (ablation) | CoDR · VQAv2 4-shot (IDEv2) | accuracy: 66.33 -> 67.01 | No clear effect |
| larger coreset beyond about 50 for IC/VQA (ablation) | CoDR · IC and VQA | CIDEr / accuracy | No clear effect |
| coreset smaller than the 120 classes (ablation) | CoDR · Stanford Dogs FIC | accuracy | Hurt |
| use similarity-based retrieval instead of random | Similarity retrieval (SITR) · MSCOCO captioning (OFv2) 1-shot | CHAIRi: 7.9 -> 14.35 | Hurt |
| use CoDR instead of infoscore-local retrieval | CoDR · MSCOCO captioning (OFv2) | time per query (s): 13.78 -> 3.95 | Improved |
| use CoDR instead of similarity-based retrieval (SITR) | CoDR · MSCOCO captioning (OFv2) | CIDEr 4-shot: 97.29 -> 104.23 | Improved |
Closest papers by meaning
- Mixture of Demonstrations for In-Context Learning (NeurIPS 2024 · similarity 0.92)
- Comparative Analysis of Demonstration Selection Algorithms for In-Context Learning in Large Language Models (Student Abstract) (AAAI 2025 · similarity 0.91)
- Large Language Models are Demonstration Pre-Selectors for Themselves (ICML 2025 · similarity 0.90)
- Provoking Multi-modal Few-Shot LVLM via Exploration-Exploitation In-Context Learning (CVPR 2025 · similarity 0.88)
- Demonstration Selection for In-Context Learning via Reinforcement Learning (ICML 2025 · similarity 0.88)
- MEND: Meta Demonstration Distillation for Efficient and Effective In-Context Learning (ICLR 2024 · 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.
