Exploring Imbalanced Annotations for Effective In-Context Learning

TL;DR

This paper investigates how class imbalance in datasets affects in-context learning with large language models and proposes a reweighting method called RCB to improve performance by addressing class and conditional biases.

Contribution

The paper introduces RCB, a novel reweighting approach that accounts for class and conditional biases, significantly enhancing ICL performance under imbalanced data conditions.

Findings

RCB improves average accuracy by up to 5.42%

Class imbalance degrades ICL performance across tasks

Class rebalancing alone is insufficient without considering conditional bias

Abstract

Large language models (LLMs) have shown impressive performance on downstream tasks through in-context learning (ICL), which heavily relies on the demonstrations selected from annotated datasets. However, these datasets often exhibit long-tailed class distributions in real-world scenarios, leading to biased demonstration selection. In this work, we show that such class imbalances significantly degrade the ICL performance across various tasks, regardless of selection methods. Moreover, classical rebalancing methods, which focus solely on class weights, yield poor performance due to neglecting condition bias--skewed feature distributions within classes. To address this, we propose Reweighting with Conditional Bias (dubbed RCB), a simple and complementary approach to enhance ICL performance under class imbalance. In particular, RCB estimates conditional bias using a balanced subset and re-weights demonstration scores based on both class weight and conditional bias. In effect, RCB prevents over-selection from dominant classes while preserving the efficacy of current selection methods. Extensive experiments on common benchmarks demonstrate the effectiveness of our method, improving the average accuracy of current selection methods by up to 5.42%.