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

Abstract

Abstract-One of the most popular feature extraction algorithms for Brain-Computer Interfaces (BCI) is Common Spatial Patterns (CSP). 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 4 new RCSP algorithms. Finally, we compare the performances of 11 different RCSP (including the 4 new ones and the original CSP), on EEG data from 17 subjects, from BCI competition data sets. 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.

An adaptive EEG filtering approach to maximize the classification accuracy in motor imagery

“…To show the effectiveness of the suggesting approach, we integrate our proposed methods on the framework proposed in [13] to keep the same environment. Table 2 reports the classification accuracies obtained on data sets for different couples of feature extraction and filters.…”

“…For the three data-sets, trials are extracted in reference to the Trigger. Based on some experiments, the epoch (window) had duration of 2 seconds for data set (IIIa, IIa) and 0.2 second for data set IVa [13]. Figure 1 depicts an example of two trials for one subject for each data set.…”

“…The main idea of this technique is to design a pair of spatial filters such that the filtered signal's variance is maximal for one class while minimum for others [1]. For comparison purpose, we use also other variants from the CSP which are SRCSP (Spatially regularized CSP), CCSP1 (Composite Common Spatial Pattern) and DLCSPauto (Diagonal Loading CSP automatic) [13]. A classifier algorithm is applied to check the classification accuracy using the selected feature.…”

“…For this range of frequencies, these data should be processed through one of the above mentioned filter techniques to keep the interesting frequency of the system in its bandwidth. In our case, the theoretical frequency response for the MI area is located on the µ-rhythm [8][9][10][11][12][13] Hz and β-rhythm [14][15][16][17][18][19][20][21][22][23][24][25][26][27][28][29][30] Hz [2]. In our case, the only known parameter is the attenuation band which should be close to zero decibel to avoid any distortion of the EEG information [16].…”

From Euclidean to Riemannian Means: Information Geometry for SSVEP Classification

“…They are efficient on clean datasets obtained from strongly constrained environment. However they are sensitive to artifacts and outliers [13,19]. Working directly on covariance matrices is advantageous: it simplifies the whole BCI system [21], avoiding the alignment of two learning steps (spatial filters and classifiers) that might lead to overfitting.…”