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?

Image: Mike Tigas from Columbia, MO, United States, CC BY 2.0, via Wikimedia Commons

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?

Imagine you're at a party with a group of friends, and everyone starts discussing the same topic after a few minutes. It's like everyone's conversation is getting influenced by the others.

In a group chat, if everyone's talking about the same thing, it's as if their conversations are blending together. This is similar to what's called "over-smoothing" in Graph Neural Networks (GNNs), where node features converge and become indistinguishable.

Example

If you have 5 friends at a party, and after 10 minutes, all of them are talking about the same movie, that's like over-smoothing happening in a GNN.

Remember this

Over-smoothing in GNNs means that after several iterations, all nodes' features become too similar, losing their unique characteristics.

Related concepts

Swipe through 100 ML concepts daily

Open Pocket Polymath