STaR: Towards Cognitive Table Reasoning via Slow-Thinking Large Language Models DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation email; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/2018/06

Huajian Zhang, Mingyue Cheng*, Yucong Luo, Xiaoyu Tao · 2025

Read STaR: Towards Cognitive Table Reasoning via Slow-Thinking Large Language Models DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation email; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/2018/06 by Huajian Zhang, Mingyue Cheng*, Yucong Luo, Xiaoyu Tao (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-2511-11233.txt.

Description

Huajian Zhang, Mingyue Cheng, and their colleagues present STaR, a slow-thinking model for reasoning over tables with large language models. Table reasoning matters for systems that have to read and analyze tabular data, but current methods reason shallowly and wobble from run to run. Shallow, implicit steps miss multi-step logic, and model uncertainty makes answers unstable.

Training runs in two stages: supervised fine-tuning on a self-verified data set, then reinforced fine-tuning with a difficulty-aware reward. A trajectory-level uncertainty score fuses token confidence with answer consistency to pick better reasoning paths. Wide experiments back the gains in depth and stability.

Catalog record

CREATORHuajian Zhang, Mingyue Cheng*, Yucong Luo, Xiaoyu Tao
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE46 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-11233

Browse the full library · About the Tower of Babel