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

被引：24

作者：

Kuznetsov, Vitaly ^{[1
]}

Mohri, Mehryar ^{[2
]}

机构：

[1] Google Res, 76 Ninth Ave, New York, NY 10011 USA

[2] Google Res & Courant Inst, 251 Mercer St, New York, NY 10012 USA

来源：

ANNALS OF MATHEMATICS AND ARTIFICIAL INTELLIGENCE | 2020年 / 88卷 / 04期

关键词：

Time series; Forecasting; Non-stationary; Non-mixing; Generalization bounds; Discrepancy; Expected sequential covering numbers; Sequential Rademacher complexity; CONVERGENCE; PREDICTION; BOUNDS;

D O I：

10.1007/s10472-019-09683-1

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

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.

引用

页码：367 / 399

页数：33

共 45 条

[1] UNIFORM CONVERGENCE OF VAPNIK-CHERVONENKIS CLASSES UNDER ERGODIC SAMPLING [J].