Discrepancy-Based Theory and Algorithms for Forecasting Non-Stationary Time Series

We present data-dependent learning bounds for the general scenario of non-stationary non-mixing stochastic processes. Our learning guarantees are expressed in terms of a data-dependent measure of sequential complexity and a discrepancy measure that can be estimated from data under some mild assumpti...

Full description

Saved in:
Bibliographic Details
Published inAnnals of mathematics and artificial intelligence Vol. 88; no. 4; pp. 367 - 399
Main Authors Kuznetsov, Vitaly, Mohri, Mehryar
Format Journal Article
LanguageEnglish
Published Cham Springer International Publishing 01.04.2020
Springer
Springer Nature B.V
Subjects
Algorithms
Analysis
Artificial Intelligence
Complex Systems
Computer Science
Forecasting
Geophysical prediction
Hypotheses
Mathematics
Random variables
Stochastic models
Stochastic processes
Time series
Non-mixing
68T01
Expected sequential covering numbers
Sequential Rademacher complexity
Discrepancy
Time series
Non-stationary
Generalization bounds
Forecasting
Online AccessGet full text

Cover

More Information
Summary:We present data-dependent learning bounds for the general scenario of non-stationary non-mixing stochastic processes. Our learning guarantees are expressed in terms of a data-dependent measure of sequential complexity and a discrepancy measure that can be estimated from data under some mild assumptions. Our learning bounds guide the design of new algorithms for non-stationary time series forecasting for which we report several favorable experimental results.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1012-2443
1573-7470
DOI:10.1007/s10472-019-09683-1