- It is the conv layer that GraphSAGE uses.
- How it works:
-
- After the node gets the features from the neighbours, it aggregates them (via any of: mean, svg, max, etc.)
-
- Next, the layer concats the aggregated neighbour features with it’s own features
CONCAT(node_features, aggregated_neighbour_features)
-
- after we have our new feature set, we apply BatchNorm or LayerNorm
-
- Notice how this is a convolutional layer, so we can apply all the techniques we use in conv nets ONTOP of graph neural nets