Logit Separability-Driven Samples and Multiple Class-Related Words Selection for Advancing In-Context Learning

TL;DR

This paper introduces logit separability as a criterion for selecting better in-context learning samples and class-related words, leading to improved large language model performance in classification tasks.

Contribution

It proposes LICL, a novel method that uses logit separability to optimize sample and label selection and incorporates multiple class-related words for enhanced instruction clarity.

Findings

Significant performance improvements across seven datasets.

Logit separability effectively assesses sample and label clarity.

Using multiple class-related words enhances label information richness.

Abstract

Effective organization of in-context learning (ICL) demonstrations is key to improving the quality of large language model (LLM) responses. To create better sample-label pairs that instruct LLM understanding, we introduce logit separability, a criterion to assess the clarity of both samples and class-related words at the logit level. This facilitates the optimization of sample and label selection, enhancing the precision of information provided in ICL demonstrations. Additionally, we find that incorporating multiple class-related words for each sample, rather than relying on a single class name, improves performance by offering a broader range of label information. Building on these insights, we propose LICL, a logit separability-based method that jointly organizes samples and integrates multiple class-related words into each sample-label pair. Evaluations across seven classification datasets show that this approach significantly improves ICL performance by providing clearer instructions and richer label information.