Learning Relational Tabular Data without Shared Features

Zhaomin Wu, Shida Wang, Ziyang Wang, Bingsheng He · 2025

Read Learning Relational Tabular Data without Shared Features by Zhaomin Wu, Shida Wang, Ziyang Wang, Bingsheng He (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-10125.txt.

Description

Zhaomin Wu and co-authors tackle a 2025 data problem: learning from tables that share no columns and hold no matched rows. Such cross-table learning could unlock vast stored data, but lining up records across tables is hard and the space of matches is huge. Their framework, called Leal, finds soft matches as it learns.

Its core bet is that well-matched data yields lower error than mismatched data. A sampling part keeps the work fast on big tables, and the team proves it can serve well. On ten test sets, their method beat rivals by up to about a quarter in predictive skill.

Catalog record

CREATORZhaomin Wu, Shida Wang, Ziyang Wang, Bingsheng He
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE44 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-10125

Browse the full library · About the Tower of Babel