Brain network embedding
Can we apply Continuous embedding to Brain network?
The answer would depend on which Brain network that we are considering because each network may have very different properties and data generation process that should be taken into account. For instance, the structural connectivity can be modeled as a generic network and we can apply various Graph embedding methods to represent it. By contrast, a functional network involves complex underlying dynamical processes and thus simply applying static network embedding methods would likely fail to capture its key properties.
The functional connectivity network is often constructed as the correlation matrix calculated from the activity (BOLD) signal measured from each voxel, which is then aggregated at the level of functional areas (nodes). Therefore, any methods that are applied to the final network would inevitably lose rich information stored in the original time series data.
The question is how can we make use of the original time series in the representation learning of the functional brain network and how can we use such representation for downstream tasks or to deepen our understanding of brain function.