TraXion: Rethinking Pre-training Frameworks for Mobility and Beyond

Shang-Ling Hsu, Mark Tenzer, Cyrus Shahabi, Khurram Shafique · 2026

Read TraXion: Rethinking Pre-training Frameworks for Mobility and Beyond by Shang-Ling Hsu, Mark Tenzer, Cyrus Shahabi, Khurram Shafique (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-2605-06906.txt.

Description

Shang-Ling Hsu, Mark Tenzer, Cyrus Shahabi, and Khurram Shafique rethink pretraining for human mobility data, whose structure language-style recipes ignore. Visits are joint events of place, time, and activity; users keep persistent signatures; and people sharing places create cross-user signal. The authors state these as three axioms for a broader class they call multi-entity spatiotemporal event streams, which also covers authentication logs and health records.

Their TraXion framework designs objectives and architecture jointly to satisfy the axioms. One checkpoint per dataset beats task-specific baselines on anomaly detection, recommendation, and prediction tasks across six mobility datasets. Applied unchanged to enterprise authentication logs and intensive care mortality prediction, the same recipe matches or beats prior specialized work.

Catalog record

CREATORShang-Ling Hsu, Mark Tenzer, Cyrus Shahabi, Khurram Shafique
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE123 KB
LICENSEhttp://creativecommons.org/licenses/by-sa/4.0/
RIGHTSCC BY SA
SOURCEarXiv

Source: https://arxiv.org/abs/2605.06906

Browse the full library · About the Tower of Babel