Sophelio ads

Next-generation fusion reactors — SPARC, Proxima Fusion, ITER successors - operate under neutron fluxes that degrade the magnetic sensors their plasma control systems depend on. Without those sensors, the standard reconstruction codes lose their primary constraints. The plasma geometry becomes invisible. That's the problem the Fusion Equilibrium Challenge is built around: can a machine learning model reconstruct the full magnetic geometry of a fusion plasma using only what a damaged reactor will still have — coil currents and Thomson-scattering profiles? The dataset is real. 9,409 experimental shots from DIII-D at General Atomics and MAST at the UK Atomic Energy Authority. 133 GB, open on Hugging Face under CC BY 4.0. No simulations. 188 participants have submitted 862 models so far — an average of 3.5 experiments per person. The leaderboard is live. Phase 1 is open through October. Top performers present at NeurIPS 2026 in December and receive named co-authorship on the peer-reviewed results paper. No plasma physics background required — just strong ML skills and an interest in a problem that actually matters. → https://lnkd.in/gXvtbfef

$500. Named co-authorship on a peer-reviewed paper. An invited talk at NeurIPS 2026 in December. That's what's on the table for top performers in the Fusion Equilibrium Challenge. 200+ participants. 1,100+ submissions. Still open. It’s not too late to make your mark → https://lnkd.in/gXvtbfef

188 people are right now teaching machines to see inside a fusion reactor. Researchers. Students. Data engineers. Freelancers. Most of them had never touched plasma physics data before this summer. We opened the Fusion Equilibrium Challenge two months ago — 9,409 real plasma shots from two tokamaks, 133 GB of experimental data, one hard inverse problem: reconstruct the magnetic geometry of a fusion plasma without any magnetic sensors. We didn't know who would show up. 862 submissions later, we do. This is what open science actually looks like. Not a lab. Not a grant committee. A global community iterating on a problem that matters — because next-generation fusion reactors will need exactly this capability to operate. The leaderboard is live. Phase 1 is still open. It’s not too late to make your mark → https://lnkd.in/gXvtbfef #FusionEnergy #MachineLearning #NeurIPS2026 #OpenScience

Two months ago, we released 133 GB of real fusion plasma data and asked a simple question: who shows up? The answer: researchers, PhD students, ML engineers, and freelancers from across six continents — most of whom had never worked with plasma physics data before. 188 participants. 862 submissions. 3.5 models per person on average. That's a community iterating on a problem it believes is solvable. The Fusion Equilibrium Challenge — accepted to the NeurIPS 2026 Competition Track — is the first open, multi-machine fusion benchmark ever released. The data comes from two operational government-run tokamaks: DIII-D (General Atomics, San Diego) and MAST (UKAEA, Culham, UK). Phase 1 is still open. Two $500 prize tracks. Co-authorship on the results paper. Invited presentation at NeurIPS in December for top teams. If you work in ML and want your models to run on something that matters — this is it. → https://lnkd.in/gXvtbfef

No fusion background required. 188 researchers, students, and engineers have already submitted 862 models on real plasma data from DIII-D and MAST. The leaderboard is live. Phase 1 closes in October. It’s still time to join → https://lnkd.in/gXvtbfef

$500. Named co-authorship on a peer-reviewed paper. An invited talk at NeurIPS 2026 in December. That's what's on the table for top performers in the Fusion Equilibrium Challenge. 188 participants. 862 submissions. Still open. It’s not too late to make your mark → https://lnkd.in/gXvtbfef