Geometric Intelligence Lab

Research lab at @UCSBEngineering revealing the geometric signatures of natural and artificial intelligence | PI: @ninamiolane

[Geometric Intelligence Lab @geometric_intel]

🌟 New review from the lab!

Discover the geometric, topological and algebraic signatures of machine learning and deep learning models.

By @naturecomputes @johmathe @mathildepapillo D. Buracas @hansenlillemark @cashewmake2 @AbbyBertics X. Pennec & @ninamiolane !

[Mathilde Papillon🦋 mathildepapillon .bsky .social @mathildepapillo]

In the 20th century, non-Euclidean geometry transformed how we model the world with pen and paper. In this century, it’s revolutionizing how we model the world with machines.

Our review on the topic is out https://t.co/zSDLIV8Ghw, led w the brilliant
@naturecomputes @johmathe

[UCSB ECE @ucsbece]

This award, supported by ICMBE2026, recognizes exceptional contributions to molecular beam epitaxy by researchers under 40. Prof. Ahmadi is a deserving honoree and we couldn't be prouder. 💛💙

#UCSB #MBE #WideBandgap #PowerElectronics #QuantumDevices #MaterialsScience

[UCSB ECE @ucsbece]

🏆Congratulations to Prof. Elaheh Ahmadi, recipient of the 2026 Young Investigator MBE Award from the International MBE Advisory Committee!

Her research on novel ultra wide bandgap devices is pushing the boundaries of high power, high frequency, and quantum applications.🔬⚡

[Geometric Intelligence Lab @geometric_intel]

Way to go Daniel!!! 😼

[Daniel Holmberg @dholmbrg]

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: https://t.co/CMd0JfSd4C

A couple highlights below 🧵

[Daniel Holmberg @dholmbrg]

• Lastly, animation of an extra long 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.

[Daniel Holmberg @dholmbrg]

• 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.

[Daniel Holmberg @dholmbrg]

• 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.

[Daniel Holmberg @dholmbrg]

• 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.

[Daniel Holmberg @dholmbrg]

• 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.

[Geometric Intelligence Lab @geometric_intel]

Ever wonder how electrical signals in your brain create consciousness? 🤔💡

Dive into this question with @ninamiolane and @claire_i_webb

[Nina Miolane 🦋 @ninamiolane.bsky.social @ninamiolane]

Can geometry help explain how neuronal signals create consciousness?🤔🌐

Loved diving into this massive question with the brilliant @claire_i_webb thanks to @longnow & @berggruenInst !

Watch our full conversation here ⬇️
@ucsbece @geometric_intel @ai_ucsb

[Nina Miolane 🦋 @ninamiolane.bsky.social @ninamiolane]

Can geometry help explain how neuronal signals create consciousness?🤔🌐

Loved diving into this massive question with the brilliant @claire_i_webb thanks to @longnow & @berggruenInst !

Watch our full conversation here ⬇️
@ucsbece @geometric_intel @ai_ucsb

[UCSB ECE @ucsbece]

Math, machine learning, and the mind💡

ECE's Prof. @ninamiolane takes a fascinating geometric approach to understanding how the brain's electronic signals give rise to consciousness.

Don't miss her conversation with @claire_i_webb! 📐🧠

[Nina Miolane 🦋 @ninamiolane.bsky.social @ninamiolane]

Fantastic idea! If you're building Graph Foundation Models, try testing them on #OgBench 🧬

By @louisacornelis @LouisVLang @johmathe @gbg1441 @geometric_intel

[Geometric Intelligence Lab @geometric_intel]

Most GNN benchmarks weren't built with biology in mind. #OgBench was. 🎉

The first benchmark for graph learning on omics data, addressing the n≪p regime across transcriptomics, proteomics, and epigenomics . ✨

Congratulations @louisacornelis @johmathe @gbg1441 @LouisVLang 🚀

[Louisa Cornelis @louisacornelis]

⁠We're releasing OgBench: the first benchmark for GNNs on omics data in the n≪p regime. Every major benchmark (OGB, TUDataset, LRGB, GraphBench) operates where the number of graphs (n) far exceeds nodes per graph (p). But biology is the opposite: few patients, massive graphs.

[Geometric Intelligence Lab @geometric_intel]

@michael_galkin @louisacornelis Great idea!

[Nina Miolane 🦋 @ninamiolane.bsky.social @ninamiolane]

Standard GNN benchmarks do not reflect the reality of omics data.

#OgBench changes that by tackling the n≪p omics regime: massive graphs of proteins/transcripts/genes, but limited patients.

A much-needed tool for AI in omics!🧬📊@louisacornelis @johmathe @gbg1441 @LouisVLang

[Michael Galkin @michael_galkin]

Thanks @louisacornelis! Hey GFM crowd, OgBench looks like a perfect example for evaluating zero-shot generalization of graph FMs without training on it (as num samples is rather small but statistically meaningful).

[Mathilde Papillon🦋 mathildepapillon .bsky .social @mathildepapillo]

@louisacornelis @LouisVLang @ninamiolane @gbg1441 @geometric_intel @johmathe A super example of why critically evaluating GNNs is so important!! Well done🙌

[Nina Miolane 🦋 @ninamiolane.bsky.social @ninamiolane]

@louisacornelis @LouisVLang @gbg1441 @geometric_intel @johmathe Fantastic collaboration with @louisacornelis @LouisVLang @johmathe @gbg1441 !

[Louisa Cornelis @louisacornelis]

Code, datasets (on HuggingFace), and a public leaderboard are all live. If you work on biological graphs, come benchmark your model and stay tuned for additional datasets.
→ https://t.co/Y0gGB2lyfl

[Louisa Cornelis @louisacornelis]

These findings echo recent warnings that graph learning risks losing relevance due to poor benchmarks. OgBench is our answer: objective evaluation to redirect attention from incremental gains to rethinking whether biological priors should be encoded as graphs at all.