@ml.swe: How do you feed a graph to a neural network? A neural network only takes numbers, but a graph is nodes and the connections between them. So every node gets a vector (a list of numbers), and each layer does two things: 1 Blend: each node averages its vector with its neighbors'. 2 Stretch: a regular neural network layer, with the same weights for every node. Repeat, and nodes from the same part of the graph end up together. In this reel the vectors start out random, so the connections are what sort them. That's a graph neural network. (The blend step is called message passing.)

Andrew Tang

Open In TikTok:

Region: US

Tuesday 06 October 2026 05:56:16 GMT