“…The reduction of the dimension of neural feature representations is mainly performed in offline or online motor BCI studies by means of projection methods, such as the principal component analysis and its variants (Devulapalli, 1996 ; Wu et al, 2003b ; Kim S.-P. et al, 2006 ; Aggarwal et al, 2008 ; Ke and Li, 2009 ; Wang W. et al, 2009 ; Argunşah and Çetin, 2010 ; Suk and Lee, 2010 ; Bhattacharyya et al, 2011 ; Kao et al, 2013 , 2017 ) or by means of feature selection methods, such as stepwise forward (Brunner et al, 2007 ; Liang and Bougrain, 2012 ; Wang et al, 2012 ; Hotson et al, 2014 ) or forward-backward (McFarland et al, 2010 ) selection procedures, LASSO-based sparse modeling methods (Least Absolute Shrinkage and Selection Operator) (Fazli et al, 2011 ; Kelly et al, 2012 ; Wang et al, 2015 ), so-called filter methods (Schalk et al, 2007 ; Spüler et al, 2016 ), genetic algorithms (Flotzinger et al, 1994 ; Graimann et al, 2004 ; Wei et al, 2006 ; Boostani et al, 2007 ; Fatourechi et al, 2007 ; Wei and Tu, 2008 ) or alternative approaches such as distinctive sensitive learning vector quantization (Flotzinger et al, 1994 ).…”