Search research articles

Search research articles

Related Experiment Videos

Learning and predicting time series by neural networks.

Ansgar Freking¹, Wolfgang Kinzel, Ido Kanter

¹Institut für Theoretische Physik und Astrophysik, Universität Würzburg, Am Hubland, 97074 Würzburg, Germany.

Physical Review. E, Statistical, Nonlinear, and Soft Matter Physics

|June 13, 2002

Summary

This summary is machine-generated.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by

Same author

Advanced deep architecture pruning using single-filter performance.

Physical review. E·2025

Same author

Towards a universal mechanism for successful deep learning.

Scientific reports·2024

Same author

Hebbian dreaming for small datasets.

Neural networks : the official journal of the International Neural Network Society·2024

Same author

Enhancing the accuracies by performing pooling decisions adjacent to the output layer.

Scientific reports·2023

Same author

Efficient shallow learning as an alternative to deep learning.

Scientific reports·2023

Same author

Learning on tree architectures outperforms a convolutional feedforward network.

Scientific reports·2023

ABOUT JoVE

Overview Leadership Blog JoVE Help Center

AUTHORS

Publishing Process Editorial Board Scope & Policies Peer Review FAQ Submit

LIBRARIANS

Testimonials Subscriptions Access Resources Library Advisory Board FAQ

RESEARCH

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments Archive

EDUCATION

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual Faculty Resource Center Faculty Site

Terms & Conditions of Use

Artificial neural networks can either predict future values in a time series or learn the underlying generating rules. This study demonstrates that these two abilities are not always linked, with implications for time series analysis.

Area of Science:

Artificial Intelligence
Machine Learning
Time Series Analysis

Background:

Artificial neural networks (ANNs) are often trained on time series data to perform two tasks: prediction and rule learning.
The relationship between prediction accuracy and the ability to learn the generating mechanism of a time series is not fully understood.

Purpose of the Study:

To investigate the dissociation between prediction and learning capabilities of ANNs trained on time series.
To determine if ANNs can simultaneously achieve accurate prediction and learn the underlying data-generating rules.

Main Methods:

Training artificial neural networks on different types of time series data.
Evaluating network performance on prediction accuracy (forecasting future values).
Assessing network performance on learning the underlying data-generating rules.

Related Experiment Videos

Main Results:

Prediction and learning are shown to be distinct abilities, not necessarily correlated.
Chaotic time series can be learned by ANNs but are difficult to predict accurately.
Quasiperiodic time series can be predicted well by ANNs but are not effectively learned.

Conclusions:

The ability of ANNs to predict a time series does not guarantee they have learned its underlying dynamics.
Different time series characteristics (e.g., chaotic vs. quasiperiodic) influence the extent to which ANNs can predict or learn.
This dissociation has significant implications for the interpretation of ANN models in time series analysis and forecasting.