Graphical models are powerful tools to express the statistical dependency of problem spaces. Accurate estimation of the graphical model
of a large set of variables has been proven to facilitate researches in computational biology, statistical physics and computer vision.
However, existing algorithms often require samples to be taken from all the variables, which is often intractable in many scenarios like brain
signal analysis where the number of electrodes plugged into the brain is limited. Therefore, we propose an active learning algorithm that
learns the structure of a high dimensional graphical model using size-constrained samples. We use simulation to demonstrate that our
algorithm could recover the graph structure correctly with sample complexity similar to those of existing algorithms using full-size samples.