Separate losses for each node for node classification

[gregoritoo]

Hi!
I would like to backpropagate different losses for each node to learn separate edge attention masks. My edge attention mask is a matrix of shape (number_of_nodes, number_of_edges) and i would like to update each row separately. Is there a simpler way to do it ?

[rusty1s]

It is not totally clear to me what you mean. If you want to learn node-specific parameters, you could think about learning an embedding matrix that is used to define edge attention scores.