Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Unsupervised learning in a BCI chess application using label proportions and expectation-maximization

Unsupervised learning in a BCI chess application using label proportions and... The online usage of brain-computer interfaces (BCI) generates unlabeled data. This data in combination with the rich structure contained in BCI applications based on event-related potentials allow to design novel unsupervised classification approaches like learning from label proportions (LLP) or its combination with expectation-maximization (EM) into a mixed model. In this work, we explore the feasibility of unsupervised classification in a BCI chess application. We propose an LLP extension based on weighted least squares regression. It requires randomization of timing parameters but overcomes the dependency on additional symbols. Simulations on electroencephalogram data obtained from six subjects playing BCI-controlled chess show that a combination of unsupervised LLP with EM (despite not using any labels) by constant adaptation quickly reaches and on the long run outperforms the average performance level of non-adaptive supervised classifiers. With our contribution, we increase the scope for which unsupervised learning methods can successfully be applied in BCI. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Brain-Computer Interfaces Taylor & Francis

Unsupervised learning in a BCI chess application using label proportions and expectation-maximization

Unsupervised learning in a BCI chess application using label proportions and expectation-maximization

Abstract

The online usage of brain-computer interfaces (BCI) generates unlabeled data. This data in combination with the rich structure contained in BCI applications based on event-related potentials allow to design novel unsupervised classification approaches like learning from label proportions (LLP) or its combination with expectation-maximization (EM) into a mixed model. In this work, we explore the feasibility of unsupervised classification in a BCI chess...
Loading next page...
 
/lp/taylor-francis/unsupervised-learning-in-a-bci-chess-application-using-label-9Z0R0xe5f3
Publisher
Taylor & Francis
Copyright
© 2020 Informa UK Limited, trading as Taylor & Francis Group
ISSN
2326-2621
eISSN
2326-263x
DOI
10.1080/2326263X.2020.1741072
Publisher site
See Article on Publisher Site

Abstract

The online usage of brain-computer interfaces (BCI) generates unlabeled data. This data in combination with the rich structure contained in BCI applications based on event-related potentials allow to design novel unsupervised classification approaches like learning from label proportions (LLP) or its combination with expectation-maximization (EM) into a mixed model. In this work, we explore the feasibility of unsupervised classification in a BCI chess application. We propose an LLP extension based on weighted least squares regression. It requires randomization of timing parameters but overcomes the dependency on additional symbols. Simulations on electroencephalogram data obtained from six subjects playing BCI-controlled chess show that a combination of unsupervised LLP with EM (despite not using any labels) by constant adaptation quickly reaches and on the long run outperforms the average performance level of non-adaptive supervised classifiers. With our contribution, we increase the scope for which unsupervised learning methods can successfully be applied in BCI.

Journal

Brain-Computer InterfacesTaylor & Francis

Published: Apr 2, 2020

Keywords: Unsupervised learning; learning from label proportions; event-related potentials; random SOA; expectation-maximization

References