New MARS method enables zero-label tabular learning using LLMs

Researchers have developed a new method called Marginal Response Surface Elicitation (MARS) for tabular learning that can make predictions without requiring labeled data. MARS leverages large language models (LLMs) to extract domain knowledge from feature semantics and task descriptions, transforming these priors into a zero-shot tabular classifier. The method involves selecting representative values for features from unlabeled data, prompting LLMs for class support scores and feature weights, and then aggregating these responses to create feature response functions for prediction. MARS has demonstrated superior performance on eight tabular benchmark tasks, achieving higher average AUC and AP scores compared to direct prompting methods while significantly reducing computational costs. AI

IMPACT Enables tabular predictions on unlabeled data, potentially reducing the need for costly data annotation in various applications.

RANK_REASON The cluster contains a research paper detailing a new method for tabular learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• alphaXiv
• Apify
• arXiv
• CatalyzeX
• DagsHub
• Gotit.pub
• Hugging Face
• Marginal Response Surface Elicitation
• MARS
• ScienceCast

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →