Can you imagine a social network where your connections influence your opinions?
message passing does in GNNs: each node aggregates features from its neighbors
Can you imagine a social network where your connections influence your opinions?
Imagine you're trying to predict the popularity of a new song by looking at the likes and shares from your friends' playlists. But everyone's playlist is different in size and order.
Think of each friend as a node in a graph, and their likes and shares as connections. Your song's popularity grows as you consider the influence of each friend's playlist, updating your prediction as you go. This is message passing.
Example
If Friend A likes the song and has 10 friends who also like it, and Friend B likes it and has 5 friends who like it, you update your popularity score by considering both Friend A and Friend B's connections.
Remember this
Message passing in GNNs allows you to update your predictions by considering the influence of your connections, similar to how your song's popularity grows with your friends' likes.
Text adapted from Wikipedia, licensed under CC BY-SA 4.0.
Related concepts
the over-smoothing problem is in GNNs: deep GNNs make all node features converge
Why do all friends in a group chat seem to start talking about the same thing after a while?
GAT (Graph Attention Network) adds: learned attention weights between neighbors
How do computers learn what to focus on in a complex network?
the lottery ticket hypothesis says: sparse subnetworks can match full network performance
Can a small part of a puzzle fit perfectly into its place by chance?
GCN (Graph Convolutional Network) does: spectral convolution approximated by neighbor averaging
How can we predict a molecule's medical efficacy?
GraphSAGE does: samples and aggregates a fixed-size neighborhood
GraphSAGE samples and aggregates a fixed-size neighborhood
Graph neural network
Graph pooling reduces graphs to single vectors for graph-level prediction
Swipe through 100 ML concepts daily
Open Pocket Polymath