Estimating brain connectivity when few data points are available: Perspectives and limitations