The Changing Brain: Bidirectional Learning Between Algorithm and User

Abstract

In 2013–2014 we have advanced our MRCP-based BCI by demonstrating: (1) the ability to detect movement intent during dynamic tasks; (2) better detection accuracy than conventional approaches by implementing the locality preserving projection (LPP) approach; (3) the ability to use a single channel for accurate detection; and (4) enhanced neuroplasticity by driving a robotic device in an online mode. To realize our final goal of an at home system, we have characterized alterations during single session use in our extracted signal when the user is undergoing complex learning or experiencing significant attentional shifts—all seriously affecting the detection of user intent. Learning enhances the variability of MRCP at specific recording sites while attention shifts result in a more global increase in signal variability. With the results presented, we are working towards an adaptive brain–computer interface where bidirectional learning (either user or algorithm) is possible.