Towards Informative Few-Shot Prompt with Maximum Information Gain for In-Context Learning
This addresses the problem of high variance in few-shot learning for LLM users, offering a more stable and effective approach, though it is incremental as it builds on existing ICL frameworks.
The paper tackles the instability in few-shot in-context learning for large language models by proposing a method to select examples with maximum information gain, achieving an average relative improvement of 14.3% across six classification tasks.
Large Language models (LLMs) possess the capability to engage In-context Learning (ICL) by leveraging a few demonstrations pertaining to a new downstream task as conditions. However, this particular learning paradigm suffers from high instability stemming from substantial variances induced by factors such as the input distribution of selected examples, their ordering, and prompt formats. In this work, we demonstrate that even when all these factors are held constant, the random selection of examples still results in high variance. Consequently, we aim to explore the informative ability of data examples by quantifying the Information Gain (IG) obtained in prediction after observing a given example candidate. Then we propose to sample those with maximum IG. Additionally, we identify the presence of template bias, which can lead to unfair evaluations of IG during the sampling process. To mitigate this bias, we introduce Calibration Before Sampling strategy. The experimental results illustrate that our proposed method can yield an average relative improvement of 14.3% across six classification tasks using three LLMs.