A Comparative Study on How Data Normalization Affects Zero-Shot Generalization in Time Series Foundation Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Ahmed, Ihab, Krompaß, Denis, Feng, Cheng, Tresp, Volker
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908795286847488
author Ahmed, Ihab
Krompaß, Denis
Feng, Cheng
Tresp, Volker
author_facet Ahmed, Ihab
Krompaß, Denis
Feng, Cheng
Tresp, Volker
contents We investigate input normalization methods for Time-Series Foundation Models (TSFMs). While normalization is well-studied in dataset-specific time-series models, it remains overlooked in TSFMs where generalization is critical. Time-series data, unlike text or images, exhibits significant scale variation across domains and channels, coupled with non-stationarity, can undermine TSFM performance regardless of architectural complexity. Through systematic evaluation across four architecturally diverse TSFMs, we empirically establish REVIN as the most efficient approach, reducing zero-shot MASE by 89\% relative to an un-normalized baseline and by 44\% versus other normalization methods, while matching the best in-domain accuracy (0.84 MASE) without any dataset-level preprocessing -- yielding the highest accuracy-efficiency trade-off. Yet its effect utilization depends on architectural design choices and optimization objective, particularly with respect to training loss scale sensitivity and model type (probabilistic, point-forecast, or LLM-based models).
format Preprint
id arxiv_https___arxiv_org_abs_2512_02833
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle A Comparative Study on How Data Normalization Affects Zero-Shot Generalization in Time Series Foundation Models
Ahmed, Ihab
Krompaß, Denis
Feng, Cheng
Tresp, Volker
Machine Learning
Artificial Intelligence
We investigate input normalization methods for Time-Series Foundation Models (TSFMs). While normalization is well-studied in dataset-specific time-series models, it remains overlooked in TSFMs where generalization is critical. Time-series data, unlike text or images, exhibits significant scale variation across domains and channels, coupled with non-stationarity, can undermine TSFM performance regardless of architectural complexity. Through systematic evaluation across four architecturally diverse TSFMs, we empirically establish REVIN as the most efficient approach, reducing zero-shot MASE by 89\% relative to an un-normalized baseline and by 44\% versus other normalization methods, while matching the best in-domain accuracy (0.84 MASE) without any dataset-level preprocessing -- yielding the highest accuracy-efficiency trade-off. Yet its effect utilization depends on architectural design choices and optimization objective, particularly with respect to training loss scale sensitivity and model type (probabilistic, point-forecast, or LLM-based models).
title A Comparative Study on How Data Normalization Affects Zero-Shot Generalization in Time Series Foundation Models
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2512.02833