When QHack rolled around in late Feb, my team and I were on the hunt for appealing QML projects. We took one glance at this quantum graph neural network paper and knew it wouldn’t disappoint. Indeed it didn’t.

5 days later, we ended up placing in the top 5 teams and winning several power-ups & prizes along the way.

This project was an exotic blend of many things I love.

Graph theory. Machine learning. Quantum Computing.

And it also sparked some new passions: high-energy physics. No less, working on this project taught me a marriage of these concepts. And so in this post, I’d like to share this exotic fusion with you. In detail.

Let me preface that my intention here is not to convince you of the quantum upside — which remains unknown — but rather to journey you through the jungle of learnings reaped from exploring this QML application.

We’ll do an in-depth breakdown of graph neural networks, how the quantum analogue differs, why one would think of applying it to high energy physics, and so much more. This post is for you if:

if you’re interested in the ins & outs of intriguing QML applications
you’re keen to get a snapshot understanding of how classical graph neural networks work
you’re the kind of person that gets dopamine hits when you make cross-domain connections between ideas

Ready to begin? Here’s the map we’ll follow.

We’ll start with understanding our problem setting at CERN, touching on what’s being done to solve it right now, and why graphML could be the potential key to solving it.
Then we’ll make an interlude to understand classical graph neural networks
So that we can break down the quantum graph neural network in detail
And end off by sharing how my group & I contributed to the existing work.

Allons-y!

Problem

Before diving right into the QML, let’s understand the problem we’re hoping to solve along with the context it resides upon. That takes us to Geneva.

CERN general operations

Diagram of the particle accelerator rings at CERN (27km in circumference) | Image credit: CERN