Hi, I'm wondering why we need a 2 layer loop in the aggregate function?
Here is the pseudo code from the paper:
From my understanding, the aggregation is a one way process, from the bottom layer(with the most nodes) to the top layer(original batch nodes), which does not need 2 loops.
Hi, I'm wondering why we need a 2 layer loop in the aggregate function? Here is the pseudo code from the paper:
From my understanding, the aggregation is a one way process, from the bottom layer(with the most nodes) to the top layer(original batch nodes), which does not need 2 loops.