Efficient machine learning algorithm for electroencephalogram modeling in brain–computer interfaces

Abstract

Brain–computer interfaces (BCIs) provide the measurement of the activities of central nervous systems, and they convert the activities into artificial outputs. Currently, one of the most interesting researches in BCIs is to develop methods for decoding people’s intent from the neural signals. Electroencephalogram (EEG) is one of the ideal solution that provides a portable recording system of neural signals. One of the challenges is to decode people’s intended movement based on EEG. Machine learning is a powerful tools and has been used in BCIs. Due to the large-scale computations, the model training process for machine learning is not always efficient. In order to address such challenges of electroencephalogram modeling in BCIs, we present an efficient machine learning algorithm based on normal equation. First, we propose a systolic matrix multiplication of two matrices. Second, we propose a systolic matrix inversion for large matrix. Third, we propose a systolic matrix-vector multiplication. In addition, the efficiency of the machine learning algorithm is analyzed and its applications of electroencephalogram modeling in BCIs are discussed.