Revisiting Demonstration Selection Strategies in In-Context Learning

TL;DR

This paper investigates the factors influencing demonstration selection in in-context learning with large language models, proposing a new data- and model-dependent method, TopK + ConE, that improves performance across tasks.

Contribution

It introduces a novel demonstration selection method, TopK + ConE, based on the correlation between demonstration contribution and model understanding, enhancing ICL effectiveness.

Findings

Consistent improvements across language understanding and generation tasks

Demonstration choice depends on data and model factors

Provides a unified explanation for previous methods' effectiveness

Abstract

Large language models (LLMs) have shown an impressive ability to perform a wide range of tasks using in-context learning (ICL), where a few examples are used to describe a task to the model. However, the performance of ICL varies significantly with the choice of demonstrations, and it is still unclear why this happens or what factors will influence its choice. In this work, we first revisit the factors contributing to this variance from both data and model aspects, and find that the choice of demonstration is both data- and model-dependent. We further proposed a data- and model-dependent demonstration selection method, \textbf{TopK + ConE}, based on the assumption that \textit{the performance of a demonstration positively correlates with its contribution to the model's understanding of the test samples}, resulting in a simple and effective recipe for ICL. Empirically, our method yields consistent improvements in both language understanding and generation tasks with different model scales. Further analyses confirm that, besides the generality and stability under different circumstances, our method provides a unified explanation for the effectiveness of previous methods. Code will be released.