iiinova — where your digital self keeps evolving

Daniel Holmberg @dholmberg.bsky.social · 01/06/2026

• Lastly, animation of an extra long sea ice concentration rollout over the spring of 2024 for the Baltic Sea, with reanalysis atmospheric forcing. The regional model remains surprisingly stable while only trained with 2 autoregressive steps.

033

Daniel Holmberg @dholmberg.bsky.social · 01/06/2026

• Sea ice is predicted alongside other physical state variables. Smooth invertible activation functions together with a binary density channel keep ice variables within realistic bounds.

132

Daniel Holmberg @dholmberg.bsky.social · 01/06/2026

• K-means cluster meshes. Latitude weighted spherical K-means produces a mesh that conforms better to ocean grids by construction compared to previously used quadrilateral or icosahedral meshes.

132

Daniel Holmberg @dholmberg.bsky.social · 01/06/2026

• Latent-variable graph neural network. A per-step Gaussian latent on the coarsest mesh level injects stochasticity into a hierarchical encode-process-decode backbone. The model only requires one forward pass per ensemble member.

142

Daniel Holmberg @dholmberg.bsky.social · 01/06/2026

• Njord is trained globally at 0.25° and on the Baltic Sea at 2km resolution. In the regional setting Njord conditions on boundary data from an independent global ocean model, where previous emulators either lack boundary forcing or depend on the very system they aim to replace.

133

Daniel Holmberg @dholmberg.bsky.social · 01/06/2026

The ocean is inherently chaotic, yet existing data-driven ocean models produce deterministic forecasts. In our new preprint, we introduce Njord, a probabilistic graph neural network for ensemble ocean forecasting. Link: arxiv.org/abs/2605.15470 A couple highlights below 🧵

263

Daniel Holmberg @dholmberg.bsky.social · 01/12/2025

We present graph-based neural surrogates that enable fast and uncertainty-aware emulation of global hybrid-Vlasov simulations via a probabilistic latent-variable model. Data + code are openly available, and compatible with advances happening in data-driven weather prediction!

042

Daniel Holmberg @dholmberg.bsky.social · 01/12/2025

Are you interested in ML for space/plasma physics? I'll be presenting “Graph-based Neural Space Weather Forecasting” this Saturday at the ML for physics workshop in San Diego @neuripsconf.bsky.social 🚀 📄 Paper: arxiv.org/abs/2509.19605 💻 Code: github.com/fmihpc/space... #AI4Science #ML4PS

183

Reposted by Daniel Holmberg

Nina Miolane @ninamiolane.bsky.social · 22/10/2025

I am recruiting PhD students for 2026!😃 You want to reveal the geometric signatures of natural and artificial intelligence, and understand computations in brains and AI? 🌐🧠🤖 Apply to the UCSB Geometric Intelligence Lab ✨ This is the view you'd have from... your desk🌴

2208

Reposted by Daniel Holmberg

Stephan Rasp @raspstephan.bsky.social · 23/12/2024

ECMWF with two new papers right before christmas. AIFS-CRPS: arxiv.org/abs/2412.158... GraphDOP (the first truly end2end global weather model): arxiv.org/abs/2412.15687 Here they are added to the SotA tracker: docs.google.com/spreadsheets...

4206

Daniel Holmberg @dholmberg.bsky.social · 08/12/2024

Hi, possible to be added here? Researching ML & ocean

110

Reposted by Daniel Holmberg

Miles Cranmer @milescranmer.bsky.social · 02/12/2024

🧵 Today with @polymathicai.bsky.social and others we're releasing two massive datasets that span dozens of fields - from bacterial growth to supernova! We want this to enable multi-disciplinary foundation model research.

38825

Reposted by Daniel Holmberg

Annalena Kofler @annalenakofler.bsky.social · 02/12/2024

1/ 🚀 New Paper Alert: Spotlight at NeurIPS ML and the Physical Sciences Workshop! We explore the intersection of high-energy physics and machine learning. What's the challenge we’re targeting, and why does it matter? Let's dive in! 🧵👇 🚀 #AI #MachineLearning #Physics #ML4PS #NeurIPS #AcademicSky

Scientific poster with dark-blue background. Title: Flow Annealed Importance Sampling Bootstrap meets differentiable particle physics. The poster contains multiple sections: Generation of data in particle physics, normalizing flows, training approaches, and results for a 2D distribution (lambda_c^+ decay) and a 8D distribution (top production). Additionally, results for the efficiency over the number of target evaluations are shown.

57816

Reposted by Daniel Holmberg

Nikolay Koldunov @oceanographer.bsky.social · 29/11/2024

🌊 Ocean currents around Antarctica from our high-resolution (3 km) ocean model. 🧪 #FESOM #SciArt

624660

Daniel Holmberg @dholmberg.bsky.social · 25/11/2024

The results show that SeaCast provides: 🔎 Accurate SST forecast with respect to satellite observations. 📊 On par skill compared to the operational forecasting system at several depth levels. 🚀 Faster predictions than a physics-based model, with a complete forecast taking 11 seconds on a single GPU.

000

Daniel Holmberg @dholmberg.bsky.social · 25/11/2024

Our model has been trained to predict physical quantities in the Mediterranean Sea up to 15 days ahead at a high 1/24° spatial resolution, using open data from Copernicus Marine Service produced by the CMCC Foundation. Both numerical and data-driven forecasts by ECMWF are tested as surface forcings.

100

Daniel Holmberg @dholmberg.bsky.social · 25/11/2024

🌊 Meet SeaCast, a new, openly available graph neural network for regional ocean forecasting. Paper: arxiv.org/abs/2410.11807. Looking forward to presenting this work at the Climate Change AI workshop taking place at @neuripsconf.bsky.social in December! 🧵 Small thread below.

171