Research map / Label-efficient and robust learning / Recognition and long-tail
Unsupervised Process-Aware Coreset Selection for In-Context Learning
Wei Zheng, Zijie Wang, Xin Li, Bin Gong, YUQING SUN
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 the representativeness term Rep(S) (ablation) | PaCS-ICL · SST-5 (budget 18, LLaMA3-8B) | accuracy: 33.5 -> 38.2 | Improved |
| add the diversity term Div(S) (ablation) | PaCS-ICL · SST-5 (budget 18, LLaMA3-8B) | accuracy: 34.1 -> 38.2 | Improved |
| add the balance factor alpha (ablation) | PaCS-ICL · SST-5 (budget 18, LLaMA3-8B) | accuracy: 34.1 -> 38.2 | Improved |
| add the representativeness term Rep(S) (ablation) | PaCS-ICL · SICK (budget 18, LLaMA3-8B) | accuracy: 55.4 -> 75.8 | Improved |
| add the diversity term Div(S) (ablation) | PaCS-ICL · SICK (budget 18, LLaMA3-8B) | accuracy: 77.7 -> 75.8 | Hurt |
| add the diversity term Div(S) (ablation) | PaCS-ICL · MRPC (budget 100, LLaMA3-8B) | accuracy: 64.8 -> 68.7 | Improved |
| add the representativeness term Rep(S) (ablation) | PaCS-ICL · SICK (budget 100, LLaMA3-8B) | accuracy: 61.8 -> 73.4 | Improved |
| add the balance factor alpha (ablation) | PaCS-ICL · MRPC (budget 100, LLaMA3-8B) | accuracy: 65.1 -> 68.7 | Improved |
| use median kNN distance radius instead of fixed r = 0.3 (ablation) | PaCS-ICL · 7-dataset average (budget 18, LLaMA3-8B) | accuracy: 61.7 -> 64.7 | Improved |
| use median radius instead of 0.25 quantile (ablation) | PaCS-ICL · 7-dataset average (budget 18, LLaMA3-8B) | accuracy: 60.3 -> 64.7 | Improved |
| use median radius instead of 0.75 quantile (ablation) | PaCS-ICL · 7-dataset average (budget 100, LLaMA3-8B) | accuracy: 61.0 -> 63.8 | Improved |
| select coreset with PaCS-ICL instead of InfoMax | coreset selection · 7-dataset average (budget 18, LLaMA3-8B) | accuracy: 61.5 -> 64.7 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- IDEAL — IDEAL: Query-Efficient Data-Free Learning from Black-Box Models (ICLR 2023) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- STAFF: Speculative Coreset Selection for Task-Specific Fine-tuning (ICLR 2025 · similarity 0.86)
- GistScore: Learning Better Representations for In-Context Example Selection with Gist Bottlenecks (ICML 2024 · similarity 0.86)
- MAPLE: Many-Shot Adaptive Pseudo-Labeling for In-Context Learning (ICML 2025 · similarity 0.86)
- Implicit In-context Learning (ICLR 2025 · similarity 0.86)
- IDEAL: Influence-Driven Selective Annotations Empower In-Context Learners in Large Language Models (ICLR 2024 · similarity 0.86)
- Large Language Models are Demonstration Pre-Selectors for Themselves (ICML 2025 · similarity 0.86)
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.
