Inter-brain Information Exploration With Applications To Brain-computer Interfaces

Under the same external stimuli, the evoked brain activities recordedby electroencephalography(EEG) show both similarity and diversity across thebrains ofdifferent subjects. Rich information hidden within the subject dimension remains to be explored.This thesis aims to explore the inter-brain information and apply this knowledge to solve the problems in the field of brain-computer interfaces(BCI).Transient evoked potentials and steady state evoked potentials are two kinds of EEG signals. Due to their different natures,distinct strategies are employed to exploit the inter-brain information in this thesis.For single-trial transient evoked potentials, since their signal-to-noise ratio is rather low and they aredominated by the spontaneous EEG, the signals across different subjects show large diversities. The information integration strategy is employed to exploit the inter-brain information in this thesis. First, by fusing multi-brain event-related potentials(ERPs), a collaborative BCIis built for fast decision making, which expands the applications of BCIs.Also, based on the diversity across subjects in a collaborative BCI,new methods forintegrating multi-brain information and upgrading classifiersare proposed.In addition, by introducing the concept of similarity-aware entropy score and multi-task learning method, this thesis further explores the similarity across subjects and proposes new methods and potential applications for multi-brain information integration.For steady state evoked potentials, since theyhave a relativelyhigh signal-to-noise ratio and they aredominated by the external stimulus, the signals across different subjects show large similarities. The information transfer strategy is employed to exploit the inter-brain information in this thesis. First,the tt-CCA(transfer template-based canonical correlation analysis) and ott-CCA(online transfer template-based CCA) algorithmswhich can transfer the information from the existing subjects’ data to a new subject are proposed to enhance the performance of BCIs based on steady state visual evoked potentials(SSVEP).These algorithms are training-free, which facilitate the use of BCIs.In addition, by further exploring the diversity across subjects, this thesis proposes the owtt-CCA(online weighted transfer template-based CCA) algorithm. This algorithm not only can be used to further enhance the performances of SSVEP BCIs, but also can be used in BCIs based on M-sequence modulated visual evoked potentials.Finally, the existing problems of information transfer rate(ITR) estimation in online BCIs are discussed. A guideline to deal with the uncertainty issues in online ITR estimation is proposed. Also, a task-oriented BCI test platform that can compare the performance of various BCI paradigms is provided to reduce the uncertainty and artifacts in online BCI performance evaluation.