The Complexity of Sequential Prediction in Dynamical Systems

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Raman, Vinod, Subedi, Unique, Tewari, Ambuj
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912409758728192
author Raman, Vinod
Subedi, Unique
Tewari, Ambuj
author_facet Raman, Vinod
Subedi, Unique
Tewari, Ambuj
contents We study the problem of learning to predict the next state of a dynamical system when the underlying evolution function is unknown. Unlike previous work, we place no parametric assumptions on the dynamical system, and study the problem from a learning theory perspective. We define new combinatorial measures and dimensions and show that they quantify the optimal mistake and regret bounds in the realizable and agnostic settings respectively. By doing so, we find that in the realizable setting, the total number of mistakes can grow according to \emph{any} increasing function of the time horizon $T$. In contrast, we show that in the agnostic setting under the commonly studied notion of Markovian regret, the only possible rates are $Θ(T)$ and $\tildeΘ(\sqrt{T})$.
format Preprint
id arxiv_https___arxiv_org_abs_2402_06614
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle The Complexity of Sequential Prediction in Dynamical Systems
Raman, Vinod
Subedi, Unique
Tewari, Ambuj
Machine Learning
We study the problem of learning to predict the next state of a dynamical system when the underlying evolution function is unknown. Unlike previous work, we place no parametric assumptions on the dynamical system, and study the problem from a learning theory perspective. We define new combinatorial measures and dimensions and show that they quantify the optimal mistake and regret bounds in the realizable and agnostic settings respectively. By doing so, we find that in the realizable setting, the total number of mistakes can grow according to \emph{any} increasing function of the time horizon $T$. In contrast, we show that in the agnostic setting under the commonly studied notion of Markovian regret, the only possible rates are $Θ(T)$ and $\tildeΘ(\sqrt{T})$.
title The Complexity of Sequential Prediction in Dynamical Systems
topic Machine Learning
url https://arxiv.org/abs/2402.06614