Objective: Brain-Computer Interface (BCI) technologies enable directcommunication between humans and computers by analyzing brain measurements,such as electroencephalography (EEG). BCI processing typically consists ofheuristically extracting features for specific tasks, limiting thegeneralizability of the BCI across tasks. Here, we asked whether we can find asingle generalized neural network architecture that can accurately classify EEGsignals in different BCI tasks. Approach: In this work we introduce EEGNet, acompact fully convolutional network for EEG-based BCIs. We compare EEGNet tothe current state-of-the-art approach across four different BCI classificationtasks: P300 visual-evoked potentials, error-related negativity responses (ERN),movement-related cortical potentials (MRCP), and sensory motor rhythms (SMR).We fit 12 different architectures, all with the same number of parameters, tostatistically control for the effect of model size versus model performance.Results: We show that one particular architecture performed on average the bestover all datasets, suggesting that a generic model can be used for a variety ofBCIs. We also show that EEGNet compares favorably to the current beststate-of-the-art approach for each dataset across all four datasets.Significance: Our findings suggest that a common simplified architecture,EEGNet, can provide robust performance across many different BCI modalities.
展开▼
