EdgeLM: Edge Demonstrations for Language Models' Table Understanding

2026-08-04

AI Summary

This work addresses the limitations of in-context learning in large language models (LLMs) for tabular understanding tasks, where conventional example retrieval based solely on query similarity often fails to provide discriminative information crucial for challenging cases. To enhance the discriminative power of contextual examples, the authors propose EdgeLM, a novel framework that introduces two complementary types of boundary evidence: (1) data boundary samples that share similar features but have different labels, and (2) model boundary samples that resemble the query and were previously misclassified by the model. Notably, EdgeLM requires neither model retraining nor task-specific customization. Extensive experiments across five data-cleaning tasks, fifteen datasets, and five mainstream LLMs demonstrate that EdgeLM consistently achieves state-of-the-art or near-optimal performance, significantly improving complex tabular understanding and confirming its effectiveness and broad applicability.

Abstract

Large language models (LLMs) perform table-centric prediction through in-context learning, making demonstration selection critical to performance. Existing retrieval methods prioritize similarity to the query, but similar demonstrations often reinforce the model's likely prediction rather than reveal the distinctions needed for difficult decisions. We propose EdgeLM, a retrieval framework that instead selects edge evidence, demonstrations that are both relevant to the query and informative about the decision boundary. EdgeLM retrieves two complementary forms of edge evidence by selecting data edges, nearby examples with different ground-truth labels, and model edges, similar examples previously misclassified by the deployed model. EdgeLM requires neither model retraining nor task-specific engineering. Across five data wrangling tasks, fifteen datasets, and five open-weight and proprietary LLMs, EdgeLM consistently achieves the best or near-best performance in every setting, while ablations show that the two forms of edge evidence provide complementary benefits. Our code and datasets are publicly available at https://github.com/soroushomidvar/EdgeLM.

Problem

• table understanding
• in-context learning
• demonstration selection
• decision boundary
• large language models

Innovation

• edge evidence
• in-context learning
• table understanding
• demonstration retrieval
• decision boundary