Non-Stationarity in the Embedding Space of Time Series Foundation Models

Jinmyeong Choi, Brad Shook, Artur Dubrawski · 2026

Read Non-Stationarity in the Embedding Space of Time Series Foundation Models by Jinmyeong Choi, Brad Shook, Artur Dubrawski (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-2604-16428.txt.

Description

Choi, Shook and Dubrawski examine how drift shows up inside the embeddings of time-series foundation models. Borrowing the quality-control view of a process leaving a stable regime, they test mean shifts, variance changes, linear trends and long-memory persistence under controlled conditions. They find that detectability in embedding space fades smoothly as shifts weaken, rather than vanishing at a sharp threshold. Different models also fail in distinct ways, warning against treating non-stationarity as a single, model-free effect.

Catalog record

CREATORJinmyeong Choi, Brad Shook, Artur Dubrawski
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE33271
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.16428

Browse the full library · About the Tower of Babel