Robust common spatial filters with a maxmin approach

Motoaki Kawanabe, Wojciech Samek, Klaus Muller, Carmen Vidaurre

Research output: Contribution to journalArticle

27 Citations (Scopus)

Abstract

Electroencephalographic signals are known to be nonstationary and easily affected by artifacts; therefore, their analysis requires methods that can deal with noise. In this work, we present a way to robustify the popular common spatial patterns (CSP) algorithm under a maxmin approach. In contrast to standard CSP that maximizes the variance ratio between two conditions based on a single estimate of the class covariance matrices, we propose to robustly compute spatial filters by maximizing the minimum variance ratio within a prefixed set of covariance matrices called the tolerance set. We show that this kind of maxmin optimization makes CSP robust to outliers and reduces its tendency to overfit. We also present a data-driven approach to construct a tolerance set that captures the variability of the covariance matrices over time and shows its ability to reduce the nonstationarity of the extracted features and significantly improve classification accuracy. We test the spatial filters derived with this approach and compare them to standard CSP and a state-of-the-art method on a real-world brain-computer interface (BCI) data set in which we expect substantial fluctuations caused by environmental differences. Finally we investigate the advantages and limitations of the maxmin approach with simulations.

Original languageEnglish
Pages (from-to)349-376
Number of pages28
JournalNeural Computation
Volume26
Issue number2
DOIs
Publication statusPublished - 2014 Jan 30

Fingerprint

Brain-Computer Interfaces
Artifacts
Noise
Spatial Pattern
Filter
Datasets
Tolerance
Real World
Simulation
Artifact
Fluctuations
Nonstationarity
Outliers
Data-driven

ASJC Scopus subject areas

  • Cognitive Neuroscience
  • Arts and Humanities (miscellaneous)

Cite this

Kawanabe, M., Samek, W., Muller, K., & Vidaurre, C. (2014). Robust common spatial filters with a maxmin approach. Neural Computation, 26(2), 349-376. https://doi.org/10.1162/NECO_a_00544

Robust common spatial filters with a maxmin approach. / Kawanabe, Motoaki; Samek, Wojciech; Muller, Klaus; Vidaurre, Carmen.

In: Neural Computation, Vol. 26, No. 2, 30.01.2014, p. 349-376.

Research output: Contribution to journalArticle

Kawanabe, M, Samek, W, Muller, K & Vidaurre, C 2014, 'Robust common spatial filters with a maxmin approach', Neural Computation, vol. 26, no. 2, pp. 349-376. https://doi.org/10.1162/NECO_a_00544
Kawanabe M, Samek W, Muller K, Vidaurre C. Robust common spatial filters with a maxmin approach. Neural Computation. 2014 Jan 30;26(2):349-376. https://doi.org/10.1162/NECO_a_00544
Kawanabe, Motoaki ; Samek, Wojciech ; Muller, Klaus ; Vidaurre, Carmen. / Robust common spatial filters with a maxmin approach. In: Neural Computation. 2014 ; Vol. 26, No. 2. pp. 349-376.
@article{3748235c8edc496684460d5acc706e7a,
title = "Robust common spatial filters with a maxmin approach",
abstract = "Electroencephalographic signals are known to be nonstationary and easily affected by artifacts; therefore, their analysis requires methods that can deal with noise. In this work, we present a way to robustify the popular common spatial patterns (CSP) algorithm under a maxmin approach. In contrast to standard CSP that maximizes the variance ratio between two conditions based on a single estimate of the class covariance matrices, we propose to robustly compute spatial filters by maximizing the minimum variance ratio within a prefixed set of covariance matrices called the tolerance set. We show that this kind of maxmin optimization makes CSP robust to outliers and reduces its tendency to overfit. We also present a data-driven approach to construct a tolerance set that captures the variability of the covariance matrices over time and shows its ability to reduce the nonstationarity of the extracted features and significantly improve classification accuracy. We test the spatial filters derived with this approach and compare them to standard CSP and a state-of-the-art method on a real-world brain-computer interface (BCI) data set in which we expect substantial fluctuations caused by environmental differences. Finally we investigate the advantages and limitations of the maxmin approach with simulations.",
author = "Motoaki Kawanabe and Wojciech Samek and Klaus Muller and Carmen Vidaurre",
year = "2014",
month = "1",
day = "30",
doi = "10.1162/NECO_a_00544",
language = "English",
volume = "26",
pages = "349--376",
journal = "Neural Computation",
issn = "0899-7667",
publisher = "MIT Press Journals",
number = "2",

}

TY - JOUR

T1 - Robust common spatial filters with a maxmin approach

AU - Kawanabe, Motoaki

AU - Samek, Wojciech

AU - Muller, Klaus

AU - Vidaurre, Carmen

PY - 2014/1/30

Y1 - 2014/1/30

N2 - Electroencephalographic signals are known to be nonstationary and easily affected by artifacts; therefore, their analysis requires methods that can deal with noise. In this work, we present a way to robustify the popular common spatial patterns (CSP) algorithm under a maxmin approach. In contrast to standard CSP that maximizes the variance ratio between two conditions based on a single estimate of the class covariance matrices, we propose to robustly compute spatial filters by maximizing the minimum variance ratio within a prefixed set of covariance matrices called the tolerance set. We show that this kind of maxmin optimization makes CSP robust to outliers and reduces its tendency to overfit. We also present a data-driven approach to construct a tolerance set that captures the variability of the covariance matrices over time and shows its ability to reduce the nonstationarity of the extracted features and significantly improve classification accuracy. We test the spatial filters derived with this approach and compare them to standard CSP and a state-of-the-art method on a real-world brain-computer interface (BCI) data set in which we expect substantial fluctuations caused by environmental differences. Finally we investigate the advantages and limitations of the maxmin approach with simulations.

AB - Electroencephalographic signals are known to be nonstationary and easily affected by artifacts; therefore, their analysis requires methods that can deal with noise. In this work, we present a way to robustify the popular common spatial patterns (CSP) algorithm under a maxmin approach. In contrast to standard CSP that maximizes the variance ratio between two conditions based on a single estimate of the class covariance matrices, we propose to robustly compute spatial filters by maximizing the minimum variance ratio within a prefixed set of covariance matrices called the tolerance set. We show that this kind of maxmin optimization makes CSP robust to outliers and reduces its tendency to overfit. We also present a data-driven approach to construct a tolerance set that captures the variability of the covariance matrices over time and shows its ability to reduce the nonstationarity of the extracted features and significantly improve classification accuracy. We test the spatial filters derived with this approach and compare them to standard CSP and a state-of-the-art method on a real-world brain-computer interface (BCI) data set in which we expect substantial fluctuations caused by environmental differences. Finally we investigate the advantages and limitations of the maxmin approach with simulations.

UR - http://www.scopus.com/inward/record.url?scp=84892998940&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84892998940&partnerID=8YFLogxK

U2 - 10.1162/NECO_a_00544

DO - 10.1162/NECO_a_00544

M3 - Article

VL - 26

SP - 349

EP - 376

JO - Neural Computation

JF - Neural Computation

SN - 0899-7667

IS - 2

ER -