Export Reviews, Discussions, Author Feedback and Meta-Reviews

The related work section will be rewritten to take this work into account. However, we believe that our method is fundamentally different than theirs as follows: they solve the following problem: given two trees, how similar are they? In contrast, we ask: given two graphs, can we count sub-graphs and smooth the counts? The smoothing is achieved by either using a pre-computed tree (graphlet kernels) or by using a data dependent tree (WL and SP kernels). Once the counts are smoothed, we discard the trees and compute the kernel via a dot product of the smoothed counts. Here are some brief comments: 1) Dependency between sub-structures: Our framework considers dependencies between sub-structures by explicitly encoding them into a DAG.