Subject-Independent Classification of Brain Signals using Skip Connections
Untapped potential for new forms of human-to-human communication can be found in the active research field of studies on the decoding of brain signals of human speech. A brain-computer interface system can be implemented using electroencephalogram signals because it poses more less clinical risk and can be acquired using portable instruments. One of the most interesting tasks for the brain-computer interface system is decoding words from the raw electroencephalogram signals. Before a brain-computer interface may be used by a new user, current electroencephalogram-based brain-computer interface research typically necessitates a subject-specific adaption stage. In contrast, the subject-independent situation is one that is highly desired since it allows a well-trained model to be applied to new users with little or no precalibration. The emphasis is on creating an efficient decoder that may be employed adaptively in subject-independent circumstances in light of this crucial characteristic. Our proposal is to explicitly apply skip connections between convolutional layers to enable the flow of mutual information between layers. To do this, we add skip connections between layers, allowing the mutual information to flow throughout the layers. The output of the encoder is then passed through the fully-connected layer to finally represent the probabilities of the 13 classes. In this study, overt speech was used to record the electroencephalogram data of 16 participants. The results show that when the skip connection is present, the classification performance improves notably.