1f9f9d8ff75205aa73ec83e543d8b571-Supplemental.pdf

We repeat the theorems presented in Sec. 3 and provide their proofs below. Inthis section we elaborate on the specific architectures that were used inour experiments inSec. In total, we have three types of architectures in our experiments, which differ in their classifier layers. Our first architecture is described in Tab. 8 and includes only a closing layer as a final step. Table 8: The architecture used for semi-and fully supervised node classification and inductive learning.