c5c3d4fe6b2cc463c7d7ecba17cc9de7-AuthorFeedback.pdf

Similarly, our framework providesrigorous evidenceto other theoretical resultsin the context13 of GNN, e.g., BN helps neural network training, skip connections avoid the problem of vanishing gradients. This isconsistent with thechoice ofGAT29 where additive attention is used.