Research map / Reinforcement learning / RL from feedback
Auto-regressive In-context Demonstration Selection
Yunzhe Qi, Sirui Chen, Jiaru Zou, Jingrui He
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 EOS perturbation lambda=0.01 (ablation) | AutoSelect · Winowhy | accuracy: 0.641 -> 0.657 | Improved |
| add EOS perturbation lambda=0.01 (ablation) | AutoSelect · AQuA | accuracy: 0.379 -> 0.395 | Improved |
| add Gaussian noise std 0.1 to rewards (ablation) | AutoSelect · Winowhy | accuracy: 0.657 -> 0.628 | Hurt |
| add Gaussian noise std 0.1 to rewards (ablation) | AutoSelect · Trec | accuracy: 0.393 -> 0.368 | Hurt |
| increase aggregation size from 3 to 10 (ablation) | AutoSelect · Trec | accuracy: 0.379 -> 0.428 | Improved |
| increase aggregation size from 3 to 10 (ablation) | AutoSelect · AQuA | accuracy: 0.383 -> 0.418 | Improved |
| lower initial temperature from 0.1 to 0.02 (ablation) | AutoSelect · Trec | accuracy: 0.393 -> 0.334 | Hurt |
| raise initial temperature from 0.1 to 0.2 (ablation) | AutoSelect · Trec | accuracy: 0.393 -> 0.371 | Hurt |
| replay buffer size 50 instead of 10 (ablation) | AutoSelect · Trec | accuracy: 0.361 -> 0.393 | Improved |
| replay buffer size 100 instead of 50 (ablation) | AutoSelect · Epistemic_reasoning | accuracy: 0.601 -> 0.575 | Hurt |
| increase validation set size from 100 to 500 (ablation) | AutoSelect · Winowhy | accuracy: 0.657 -> 0.635 | No clear effect |
| use auto-regressive AutoSelect instead of top-k Qwen3-8B retrieval | Qwen3-8B · Trec | accuracy: 0.412 -> 0.483 | 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.
- CEIL — CEIL: Generalized Contextual Imitation Learning (NeurIPS 2023) · better on 9 of 9 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
- Comparative Analysis of Demonstration Selection Algorithms for In-Context Learning in Large Language Models (Student Abstract) (AAAI 2025 · similarity 0.93)
- Large Language Models are Demonstration Pre-Selectors for Themselves (ICML 2025 · similarity 0.91)
- Demonstration Selection for In-Context Learning via Reinforcement Learning (ICML 2025 · similarity 0.89)
- Mixture of Demonstrations for In-Context Learning (NeurIPS 2024 · similarity 0.89)
- Sample Efficient Demonstration Selection for In-Context Learning (ICML 2025 · similarity 0.88)
- MEND: Meta Demonstration Distillation for Efficient and Effective In-Context Learning (ICLR 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.
