Online learning in neural decoding using incremental linear discriminant analysis

Neural decoding focuses on predicting behavior variables based on neural activities. Linear discriminant analysis (LDA) has been successfully used in pattern recognition and machine learning to find the set of discriminant vectors to characterize two or more classes of objects. However, LDA cannot be directly used for real-time neural decoding problems. In this paper, we propose an incremental LDA with online learning method to overcome this limitation. The dataflow techniques are implemented in the LIDE (LIghtweight Dataflow Environment), which provides capabilities to systematically optimize and integrate embedded software components for signal and information processing. Using these techniques along with online learning, an efficient real-time neural decoding system can be attained.