Fig. 4

Assignment kernels. Illustration of optimal assignment kernels with vertex embeddings. The vertices of two different graphs (left), G and H are embedded in a common space \(\mathcal {R}\) (middle). For example, vertices u∈V_G,v∈V_H are given embeddings \(x_{u}, y_{v} \in \mathcal {R}\). Finally, a bipartite graph with weights determined by the distances between the vertex embeddings of the two graphs is constructed and used to compute an optimal matching between the vertex sets. The weight of the matching is used to compute the kernel value k(G,H)