Graph neural network - long-range interaction modeling

GraphHSCN

GraphHSCN was an undergraduate research prototype for modeling relationships that span beyond local graph neighborhoods.

Graph learning Long-range structure Message passing Research prototype

Visit project

Graph neural network clusters with long-range connections and layered message passing

Question

How can a graph model reason across distant regions when important dependencies are not captured by local neighborhoods alone?

Build

A graph neural network architecture prototype focused on long-range interaction modeling and structured propagation.

Lesson

The project trained my eye for representation: what gets connected, what gets compressed, and what structure a model can actually use.

Why it belongs here

GraphHSCN predates my current HCI focus, but it is part of the path into how I think now. It made me care about structure: not only as math, but as a design choice that decides which relationships are easy or hard to express.

What I learned

Working on graph architecture helped me see modeling as a form of framing. Before a system can reason well, someone has already decided what the units are, which connections matter, and how information should move.