English: Node representations update in a Message Passing Neural Network (MPNN) layer. Node ${\displaystyle \mathbf {x} _{0}}$ receives messages from all of his immediate neighbours ${\displaystyle \mathbf {x} _{1}}$ to ${\displaystyle \mathbf {x} _{4}}$. Messages are computing via the message function ${\displaystyle \psi }$, which accounts for the features of both endpoints of an adjacency.