Embedding Foundation Model Predictions in Discrete-Choice Models with Structural Guarantees

Yingshuo Wang, Xian Sun, Yanhang Li, Zhichao Fan, Zexin Zhuang · 2026

Read Embedding Foundation Model Predictions in Discrete-Choice Models with Structural Guarantees by Yingshuo Wang, Xian Sun, Yanhang Li, Zhichao Fan, Zexin Zhuang (2026) 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-2606-26432.txt.

Description

Yingshuo Wang and four colleagues fix a clash between powerful prediction models and economic common sense in discrete-choice models, which estimate how people pick among options such as travel modes. Tabular foundation models, large pre-trained systems for table data, predict choices well but can say demand rises when price rises, or assign chances to options that do not exist. Planners cannot use numbers that break basic logic.

Their two-stage adapter places the foundation model's predicted probabilities inside a classic logit model's utility, a numerical score for each option's appeal. First the structural coefficients are fitted with sign rules enforced, then frozen while a small neural correction learns from the model predictions. The authors prove the blend keeps the logit's value-of-time formula exact. Tests gained accuracy over the plain logit while keeping cost behaviour sensible on every case.

Catalog record

CREATORYingshuo Wang, Xian Sun, Yanhang Li, Zhichao Fan, Zexin Zhuang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE54 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.26432

Browse the full library · About the Tower of Babel