Recent developments in the application of electroencephalography (EEG) signal-based brain-machine interfaces (BMI) provide support for the capabilities of EEG techniques to account for neural dynamics associated with simple task performance. However, the fundamental question remains as to whether EEG signal patterns relfect information sufficient for dictating underlying cognitive processes. Accurate identification imposes a substantial challenge to computation because such congnitive processes are known to involve a brain-wide correlation in both a spatial and temporal domain. Here we discuss a flexible computational framework for efficiently analyzing dynamics of the whole brain network. The proposed method reduces a heavy computational load by switching between covariance and gram matrices to compute eigenvectors, potentially enabling us to streamline analyses for revealing information pertaining to the present cognitive state.