Dynamic Demonstrations Controller for In-Context Learning

TL;DR

This paper introduces D$^2$Controller, a dynamic method to optimize the number of demonstrations in in-context learning, leading to significant performance improvements across various language models and datasets.

Contribution

It challenges the assumption that more demonstrations always improve ICL performance and proposes a novel dynamic controller to adaptively select demonstrations.

Findings

D$^2$Controller improves ICL performance by 4.6% across multiple LLMs.

Increasing demonstrations does not always enhance performance, contrary to common belief.

The method achieves competitive results when extended to previous ICL models.

Abstract

In-context learning (ICL) is a new paradigm for natural language processing (NLP), where a large language model (LLM) observes a small number of demonstrations and a test instance as its input, and directly makes predictions without updating model parameters. Previous studies have revealed that ICL is sensitive to the selection and the ordering of demonstrations. However, there are few studies regarding the impact of the demonstration number on the ICL performance within a limited input length of LLM, because it is commonly believed that the number of demonstrations is positively correlated with model performance. In this paper, we found this conclusion does not always hold true. Through pilot experiments, we discover that increasing the number of demonstrations does not necessarily lead to improved performance. Building upon this insight, we propose a Dynamic Demonstrations Controller (D$^2$Controller), which can improve the ICL performance by adjusting the number of demonstrations dynamically. The experimental results show that D$^2$Controller yields a 4.6% relative improvement on ten different sizes of LLMs across ten datasets. Moreover, we also extend our method to previous ICL models and achieve competitive results.