Transformers and Their Roles as Time Series Foundation Models

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Wu, Dennis, He, Yihan, Cao, Yuan, Fan, Jianqing, Liu, Han
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866913679567486976
author Wu, Dennis
He, Yihan
Cao, Yuan
Fan, Jianqing
Liu, Han
author_facet Wu, Dennis
He, Yihan
Cao, Yuan
Fan, Jianqing
Liu, Han
contents We give a comprehensive analysis of transformers as time series foundation models, focusing on their approximation and generalization capabilities. First, we demonstrate that there exist transformers that fit an autoregressive model on input univariate time series via gradient descent. We then analyze MOIRAI, a multivariate time series foundation model capable of handling an arbitrary number of covariates. We prove that it is capable of automatically fitting autoregressive models with an arbitrary number of covariates, offering insights into its design and empirical success. For generalization, we establish bounds for pretraining when the data satisfies Dobrushin's condition. Experiments support our theoretical findings, highlighting the efficacy of transformers as time series foundation models.
format Preprint
id arxiv_https___arxiv_org_abs_2502_03383
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Transformers and Their Roles as Time Series Foundation Models
Wu, Dennis
He, Yihan
Cao, Yuan
Fan, Jianqing
Liu, Han
Machine Learning
Artificial Intelligence
We give a comprehensive analysis of transformers as time series foundation models, focusing on their approximation and generalization capabilities. First, we demonstrate that there exist transformers that fit an autoregressive model on input univariate time series via gradient descent. We then analyze MOIRAI, a multivariate time series foundation model capable of handling an arbitrary number of covariates. We prove that it is capable of automatically fitting autoregressive models with an arbitrary number of covariates, offering insights into its design and empirical success. For generalization, we establish bounds for pretraining when the data satisfies Dobrushin's condition. Experiments support our theoretical findings, highlighting the efficacy of transformers as time series foundation models.
title Transformers and Their Roles as Time Series Foundation Models
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2502.03383