At Proxima Fusion, we believe commercial fusion power will transform the global energy landscape. Getting there, however, requires tackling one of the world’s most complex technical challenges: stellarator optimization.

We believe machine learning (ML) has an outsized role to play in solving that challenge, and we need the best brains on the task! 

That’s why we’re teaming up with Hugging Face to launch a collaborative challenge inviting the global machine learning community to help us solve three different stellarator design problems. The goal? Empower as many people as possible to accelerate progress toward bringing stellarator fusion power plants to the grid.

The Mission: Help Design a Fusion Machine That Works

Stellarators confine plasma using intricate magnetic fields generated by external coils. Unlike tokamaks, they promise steady-state operation and disruption-free performance – critical traits for a safe and viable fusion power plant. 

But no one ever said designing stellarators was easy – design complexity has actually been the single challenge that has held stellarators back since people started researching them in the 1950s. You need to optimize complex 3D shapes in high-dimensional space, balancing multiple physics and engineering constraints. 

That’s where you, the ML community, come in.

The Challenge: Three Stellarator Optimization Problems

The ConStellaration Challenge is an invitation to apply optimization techniques and scientific machine learning to one of humanity’s most ambitious missions.

We propose three benchmark problems of increasing complexity, each with progressive relevance to fusion reactor design:

1. Geometric Optimized Stellarator – Design a shape that minimizes elongation under fixed constraints on aspect ratio, triangularity, and magnetic field twist.
   
   
2. Simple-to-Build Quasi-Isodynamic (QI) Stellarator – Optimize for plasma shapes that promote good confinement via the (QI) property, but are simpler to build, with smoother magnetic fields requiring simpler coils.
   
   
3. Multi-Objective, MHD-Stable QI Stellarator – Map the compactness and simplicity tradeoff while ensuring confinement and stability. 

Each benchmark problem comes with reference implementations, evaluation scripts, and strong baselines using classical optimization methods. Your job is to beat them – or to propose novel methods that unlock the configuration design space in new ways.

The Tools: Open-Source Code and Physics-Powered Metrics

To support challenge participants, we’re releasing:

• An new open-source dataset of ~160,000 QI-like stellarator plasma boundaries, enabling ML researchers to create optimal stellarator configurations with novel, data-driven methods.
• VMEC++-based evaluation tools to compute performance metrics for candidate plasma shapes.
• Benchmark problems and baselines.
• A newly-published preprint outlining and expanding on all of the above.

All submissions will be evaluated using a physics-based forward model. You don’t need to be a fusion physicist – we’ve abstracted the complexity so you can focus on what you do best: (data-driven) optimization.

How the Challenge Works

• Submit plasma boundary shapes (represented by a truncated fourier series as specified in our preprint).
• The Hugging Face leaderboard computes metrics and ranks your submission allowing us all to compare it to others and our baselines.

Whether you’re interested in generative models, constrained optimization, or representation learning, there’s a way to contribute, and we want to hear from you. 

Why This Matters

With open-source tools, standardized benchmarks, and a powerful dataset, we can make stellarator optimization a machine learning problem – and potentially unlock a leap forward in commercial viability for fusion.

This is more than just a technical challenge: it’s a call to apply cutting-edge ML to a new frontier in physics and engineering. You could help shape the configuration of the world’s first stellarator fusion power plant.

So if you're ready to contribute, clone the ConStellaration dataset on Hugging Face and let’s get started on building the future together.