We present a new pattern recognition framework for Brain-Computer Interfacing that learns discriminative brain activity patterns, compact modeling, and robustness against signal variabilities by a single joint optimization. We present an algorithm based on the Alternating Direction Method of Multipliers, which finds an optimal solution for this approach extremely efficiently.
A first evaluation using a publicly available EEG motor imagery data corpus with 105 subjects shows that our framework outperformed state-of-the-art methods and successfully performed subject transfer.