#graph-neural-networks

Question 1: Single choice

A message-passing GNN aggregates the neighbour features of node $v$ as $f\big(\sum_{u \in N(v)} g(h_u)\big)$ , where $g$ is applied to each neighbour separately and $f$ sees only the sum. Which choice of $f$ and $g$ gives the mean of the neighbour features?

$g(h) = (h, 1)$ and $f(s, c) = s / c$ 1
$g(h) = h / \lvert N(v) \rvert$ and $f(s) = s$ 2
$g(h) = h$ and $f(s) = s / d$ , with $d$ the feature dimension3
$g(h) = e^{h}$ and $f(s) = \log s$ 4

#graph-neural-networks

Timed test