Figure 2 - uploaded by Konstantinos Barmpas
Content may be subject to copyright.
Dynamic convolution framework for BCI architectures. X represents input EEG signals, Y the associated MI labels. K different subjects in the training set are represented by different colors in the convolutional blocks. Colored rectangles and arrows (namely green, red and dark blue) demonstrate the different blocks that are taken into account when computing the final convolutional blocks for the MI classification task.
Source publication
In this work, we employ causal reasoning to breakdown and analyze important challenges of the decoding of Motor-Imagery (MI) electroencephalography (EEG) signals. Furthermore, we present a framework consisting of dynamic convolu-tions, that address one of the issues that arises through this causal investigation, namely the subject distribution shif...
Context in source publication
Context 1
... of having a BCI architecture that tries to discover a common latent space for all k subjects in the training set, we use k parallel trainable convolutional kernels (corresponding to the k available training subjects) for each convolutional block of a CNN-based BCI network. Using a subject attention network that learns to distinguish between the available individuals, we decouple the subjects and essentially train simultaneously k parallel personalized models of the same BCI architecture, as illustrated in Figure 2. ...
