Regularizing Common Spatial Patterns to Improve BCI Designs: Unified Theory and New Algorithms

By Fabien Lotte, Cuntai Guan. IEEE Transactions on Biomedical Engineering, 2011

1,030 citations recorded by the source corpus.

One of the most popular feature extraction algorithms for brain-computer interfaces (BCI) is common spatial patterns (CSPs). Despite its known efficiency and widespread use, CSP is also known to be very sensitive to noise and prone to overfitting. To address this issue, it has been recently proposed to regularize CSP. In this paper, we present a simple and unifying theoretical framework to design such a regularized CSP (RCSP). We then present a review of existing RCSP algorithms and describe how to cast them in this framework. We also propose four new RCSP algorithms. Finally, we compare the performances of 11 different RCSP (including the four new ones and the original CSP), on electroencephalography data from 17 subjects, from BCI competition datasets. Results showed that the best RCSP methods can outperform CSP by nearly 10% in median classification accuracy and lead to more neurophysiologically relevant spatial filters. They also enable us to perform efficient subject-to-subject transfer. Overall, the best RCSP algorithms were CSP with Tikhonov regularization and weighted Tikhonov regularization, both proposed in this paper.

Subfields: Cognitive Neuroscience, Signal Processing

View Paper: https://www.semanticscholar.org/paper/c441c937cd08f8e85e530c0eeb875730462607f5

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Cognitive Neuroscience

In Signal Processing

Citations in the corpus

BCILAB: a platform for brain–computer interface development (2013, methodology): “For comparability our analysis of this task largely replicates that of [45] up to minor differences in analysis parameters.”
Online and Offline Domain Adaptation for Reducing BCI Calibration Effort (2017, methodology): “For example, prior information on which EEG channels are the most likely to be useful was used in [28] as a regularizer to optimize spatial filters, and beamforming has been used in [16] to find relevant features from prior regions of interest to reduce the calibration data requirement.”
Transfer Learning: A Riemannian Geometry Framework With Applications to Brain–Computer Interfaces (2018, background): “Some examples of transfer learning applied to BCI problem can be found in [15], [21], [27] and [36].”
EEG-Based Spatio–Temporal Convolutional Neural Network for Driver Fatigue Evaluation (2019, positive): “A regularized CSPs algorithm was presented as feature extraction in [20], which significantly outperformed traditional CSPs in motor imagery classification.”
Learning Temporal Information for Brain-Computer Interface Using Convolutional Neural Networks (2018, background): “This representation of the high-dimensional EEG data can be easily fed into a linear classifier, such as support vector machine (SVM), leading to good performance [14].”
Transfer Learning in Brain-Computer Interfaces (2015, background): “[17], [18], Lotte and Guan [19], and Devlaminck et al.”
Signal Processing Approaches to Minimize or Suppress Calibration Time in Oscillatory Activity-Based Brain–Computer Interfaces (2015, positive): “Similarly, one can use as prior information which channels are the most likely to be useful and use that as a regularizer to optimize spatial filters [55].”
Progress in Brain Computer Interface: Challenges and Opportunities (2019, background): “In case of small training trials, regularizing the covariance estimation works better than the traditional algorithm [115].”
EEGNet: a compact convolutional neural network for EEG-based brain–computer interfaces (2016, methodology): “…user-independent model for SMR-based BCI, which was evaluated on the BCI Competition IV competition dataset and compared against 11 other approaches, is the approach outlined in [77], wherein all available subjects are pooled to train CSP ﬁlters and a linear discriminant analysis (LDA) classiﬁer.”
Review of the BCI Competition IV (2012, methodology): “The success of the CSP-based methods in the BCI competitions may have a promoting factor for the flourishing development of variants of CSP analysis (Lotte and Guan, 2011; Nikulin et al., 2011; Sannelli et al., 2011).”
HS-CNN: a CNN with hybrid convolution scale for EEG motor imagery classification (2020, background): “Guan [62] Raza et al [63] Raza et al [63] Gaur et al [64] Gaur”