Why Cutting Trajectories Into Small Pieces Helps to Learn Dynamical Systems Better: A Seemingly Counterintuitive Empirical Result Explained

Authors

Olga Kosheleva, The University of Texas at El PasoFollow
Vladik Kreinovich, The University of Texas at El PasoFollow

Publication Date

8-2020

Comments

Technical Report: UTEP-CS-20-88

Published in Applied Mathematical Sciences, 2020, Vol. 14, No. 13, pp. 653-658.

Abstract

In general, the more information we use in machine learning, the more accurate predictions we get. However, recently, it was observed that for prediction of the behavior of dynamical systems, the opposite effect happens: when we replace the original trajectories with shorter pieces -- thus ignoring the information about the system's long-term behavior -- the accuracy of machine learning predictions actually increases. In this paper, we provide an explanation for this seemingly counterintuitive result.