Selection of motor imageries for brain-computer interfaces based on partial kullback-leibler information measure

Taro Shibanoki, Yuki Koizumi, Bi Adriel Yozan, Toshio Tsuji

Research output: Chapter in Book/Report/Conference proceedingConference contribution

1 Citation (Scopus)

Abstract

This paper proposes a selection method of motor imageries for brain-computer interfaces based on partial Kullback-Leibler information measure. In this method, partial KL information is defined as ratio of before:after class elimination and can be obtained by a KL information-based probabilistic neural network training. Therefore, optimal classes can be selected by eliminating ineffective ones one at a time along with network training. In the experiments performed, various motor imageries were learned by the reduced-dimensional recurrent probabilistic neural network and quasi-optimal combinations were selected using the proposed method. The discrimination rates before and after selections were $\mathbf{19.57}\pm \mathbf{7.09}[\%]$ and $\mathbf{68.14}\pm \mathbf{21.70} [\%]$, respectively.

Original languageEnglish
Title of host publication2018 IEEE Life Sciences Conference, LSC 2018
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages243-246
Number of pages4
ISBN (Electronic)9781538667095
DOIs
Publication statusPublished - Dec 10 2018
Externally publishedYes
Event2018 IEEE Life Sciences Conference, LSC 2018 - Montreal, Canada
Duration: Oct 28 2018Oct 30 2018

Publication series

Name2018 IEEE Life Sciences Conference, LSC 2018

Conference

Conference2018 IEEE Life Sciences Conference, LSC 2018
Country/TerritoryCanada
CityMontreal
Period10/28/1810/30/18

Keywords

  • Brain Computer Interface
  • Class Selection
  • Electroencephalogram (EEG)
  • Kullback-Leibler Divergence
  • Motor Imagery

ASJC Scopus subject areas

  • Signal Processing
  • Medicine (miscellaneous)
  • Health Informatics
  • Instrumentation

Fingerprint

Dive into the research topics of 'Selection of motor imageries for brain-computer interfaces based on partial kullback-leibler information measure'. Together they form a unique fingerprint.

Cite this