Effective Demonstration Annotation for In-Context Learning via Language Model-Based Determinantal Point Process
About
In-context learning (ICL) is a few-shot learning paradigm that involves learning mappings through input-output pairs and appropriately applying them to new instances. Despite the remarkable ICL capabilities demonstrated by Large Language Models (LLMs), existing works are highly dependent on large-scale labeled support sets, not always feasible in practical scenarios. To refine this approach, we focus primarily on an innovative selective annotation mechanism, which precedes the standard demonstration retrieval. We introduce the Language Model-based Determinant Point Process (LM-DPP) that simultaneously considers the uncertainty and diversity of unlabeled instances for optimal selection. Consequently, this yields a subset for annotation that strikes a trade-off between the two factors. We apply LM-DPP to various language models, including GPT-J, LlaMA, and GPT-3. Experimental results on 9 NLU and 2 Generation datasets demonstrate that LM-DPP can effectively select canonical examples. Further analysis reveals that LLMs benefit most significantly from subsets that are both low uncertainty and high diversity.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Disease Classification | MIMIC-CXR | F1 Score0.81 | 19 | |
| Classification | PIFIR | Accuracy73.8 | 7 | |
| Medical Image Classification | PIFIR | Accuracy57.1 | 7 | |
| Medical Image Classification | CHIFIR | Accuracy90.4 | 7 | |
| Medical Report Classification | PIFIR (test) | Accuracy73.8 | 7 | |
| Medical Report Classification | CHIFIR (test) | Accuracy15.4 | 7 | |
| Transfer Learning Gap Analysis | CHIFIR to PIFIR | Delta F1 Score0.007 | 6 | |
| Transfer Gap Analysis | MIMIC-CXR to PIFIR | Delta F1 Score-0.021 | 6 |