Graph Convolutional Network by Hand: 12-Step Walkthrough

ProfTomYeh · x · 2026-08-23

Prof. Tom Yeh provides a 12-step hand-calculated walkthrough of Graph Convolutional Networks (GCNs). The demo covers adjacency matrices, self-loops, message passing, neighbor pooling, and classification. The key takeaway is that a GCN layer consists of pooling neighbors via an adjacency matrix and transforming nodes via an MLP, closely resembling a Transformer layer where the attention matrix is replaced by a fixed adjacency matrix.

Original post →

More from Research

Research channel →