224
D. Zhang
or mental states, they are not informative about the underlying neural mechanism
by themselves, i.e. lacking explanatory power [55]. Nevertheless, machine-learning
methods are advantageous when mining vast amounts of high-dimensional data, as
in the case of multi-brain analysis [33, 46].
Researchers are starting to take the machine-learning approach for analyzing
multi-brain EEG data. Movement directions can be predicted with accuracies from
66% in a single-brain condition to 95% when data from 20 brains were included
[66]. Enhanced perceptual decision accuracy was achieved by aggregating EEG
activities from a group of participants, for both a discrimination task [15] and a
visual search task [64]. In a more interactive scenario, frontal alpha oscillations
could effectively distinguish between leaders from followers involved in a motor
coordination task [34]. More studies in this direction are expected to emerge in the
near future, translating the already reported hyperlink findings into predictive power
and thus facilitating hyperlink-based applications, e.g., evaluating and training of
social interaction abilities.
10.4 Future Perspectives
The development of hyperscanning methods is at its early beginning. While the hyperscanning technique is bringing exciting findings to the field of social neuroscience,
critical methodological challenges remain to be further addressed, as summarized
below.
First, EEG recordings in naturalistic social interaction environment will inevitably
be affected by extensive artifacts due to necessary movements of the eyes, faces, limbs
etc., as well as electrical noises in the normally unshielded environment. Therefore,
artifact rejection need to be treated with high priority and caution need to be taken
when validating its effectiveness. Whereas conventional artifact rejection methods
targeting at modelling environmental or physiological noises can be readily applied,
more advanced methods are needed to be developed, in order to better remove the
possibly stronger artifacts during social interactions. For instance, it has recently
been demonstrated that artifact reduction could be achieved by modelling the valid
signals [5, 6, 11, 42].
Second, a new methodological framework need to be defined in order to deal
with data coming from different brains. Most of the EEG signal processing methods to date are based on the assumption that the neural signals are generated by
the same system (i.e. brain). Although progress has been to address this issue by
normalization of the individual brain’s data, or extraction of non-individual-specific
information [2], these latest methods work mainly within the signal space, on the
basis of temporal, spectral or spatial features. In other words, the reported hyperlinks
imply similarities across individuals in the EEG signal space. Possible similarity
in higher cognitive levels, has not been systematically investigated. Nevertheless,
the success of representational similarity analysis for visual objection recognition in
fMRI provide strong evidence for the existence of such across-individual similarity
D. Zhang
or mental states, they are not informative about the underlying neural mechanism
by themselves, i.e. lacking explanatory power [55]. Nevertheless, machine-learning
methods are advantageous when mining vast amounts of high-dimensional data, as
in the case of multi-brain analysis [33, 46].
Researchers are starting to take the machine-learning approach for analyzing
multi-brain EEG data. Movement directions can be predicted with accuracies from
66% in a single-brain condition to 95% when data from 20 brains were included
[66]. Enhanced perceptual decision accuracy was achieved by aggregating EEG
activities from a group of participants, for both a discrimination task [15] and a
visual search task [64]. In a more interactive scenario, frontal alpha oscillations
could effectively distinguish between leaders from followers involved in a motor
coordination task [34]. More studies in this direction are expected to emerge in the
near future, translating the already reported hyperlink findings into predictive power
and thus facilitating hyperlink-based applications, e.g., evaluating and training of
social interaction abilities.
10.4 Future Perspectives
The development of hyperscanning methods is at its early beginning. While the hyperscanning technique is bringing exciting findings to the field of social neuroscience,
critical methodological challenges remain to be further addressed, as summarized
below.
First, EEG recordings in naturalistic social interaction environment will inevitably
be affected by extensive artifacts due to necessary movements of the eyes, faces, limbs
etc., as well as electrical noises in the normally unshielded environment. Therefore,
artifact rejection need to be treated with high priority and caution need to be taken
when validating its effectiveness. Whereas conventional artifact rejection methods
targeting at modelling environmental or physiological noises can be readily applied,
more advanced methods are needed to be developed, in order to better remove the
possibly stronger artifacts during social interactions. For instance, it has recently
been demonstrated that artifact reduction could be achieved by modelling the valid
signals [5, 6, 11, 42].
Second, a new methodological framework need to be defined in order to deal
with data coming from different brains. Most of the EEG signal processing methods to date are based on the assumption that the neural signals are generated by
the same system (i.e. brain). Although progress has been to address this issue by
normalization of the individual brain’s data, or extraction of non-individual-specific
information [2], these latest methods work mainly within the signal space, on the
basis of temporal, spectral or spatial features. In other words, the reported hyperlinks
imply similarities across individuals in the EEG signal space. Possible similarity
in higher cognitive levels, has not been systematically investigated. Nevertheless,
the success of representational similarity analysis for visual objection recognition in
fMRI provide strong evidence for the existence of such across-individual similarity