Playground experiment

Graph Signal Diffusion

Click a node to inject a signal. Nearby nodes receive a weaker version of it at each hop. It is a small visual intuition for graph neighborhoods, message passing, and network effects.

Interactive graph

Inject a signal

Click any node in the canvas. Increase the hop count to let the signal travel farther. Increase the decay to make the signal fade more slowly.

Ready

What this represents

This is not meant to be a full GNN implementation. It shows the core intuition: a node starts with information, then its local neighborhood receives progressively weaker traces of it.

In a real graph neural network, each node repeatedly aggregates information from its neighbors and updates its representation. More hops means a wider receptive field, but also more smoothing.

The follow-up GNN toy makes that update step explicit with editable node features and one to three message-passing layers.

A shared graph language

The same visual metaphor connects several interests: protein interaction networks, disease modules, graph neural networks, random walks, music harmony graphs, and even idea association.

Keeping it small makes the interaction reusable: a graph can be a model, a map, a composition tool, or a way to make abstract relationships easier to inspect.