The paper proposes a new active learning method that learns on training graphs a graph policy network, which is applied to query the labeled nodes of a test graph. Initially, there were serious concerns about the clarity of technical details, in particular, how "zero-shot transfer learning" on node classification was defined and realized. The AC shared similar concerns and had reached to the authors via CMT to ask for additional response. The additional response from the authors, which described in detail the techincal details on how the policy network was applied to a test graph, had helped address these concerns. Consequently, a general consensus of the reviewers was reached on accepting the paper (R1 commented during the discussion: "I agree that maybe more explanations about the transferability of the model could have been elaborated in the main body of the paper, but I do think this is an interesting paper and its overall quality is good. So I am leaning towards a weak accept"; R3 commented to the AC: "I raised the score from 3 to 5, just to be fair to other submissions, which may not have the opportunity to be discussed seriously with another chance to further clarify their papers."). The AC encourages the authors to incorporate their additional response sent via CMT to their final revision, to clearly enhance the clarity of the technical details, in particular, how the graph policy network is applied to query labels on a test graph.