Julian Suk

cheers for the tip!

Maths YouTube is really taking off lately! Check out these creators that I particularly enjoy:

• 3Blue1Brown (obviously)
• Mutual Information (machine learning and computation)
• Aleph 0 (pure mathematics, beautiful visuals)
• Sheafification of G (chaotic and stimulating, lots of memes)

Paper:
• sciencedirect.com/science/article/pii/S016926072500375X
• arxiv.org/abs/2410.11920

Code:
• github.com/sukjulian/deep-vectorised-operators

Collaboration with Guido Nannini, Patryk Rygiel, @christophbrune.bsky.social, Gianluca Pontone, Alberto Redaelli and @jelmerwolterink.bsky.social

We propose a time-efficient surrogate model for the extrapolation of transient fluid dynamics from a steady-state flow and pressure field in coronary arteries. Like all neural operators, the model learns a discretisation-independent map between (infinite-dimensional) function spaces.

Neural operators are gaining traction in science and engineering as fast surrogates for time-intensive physics simulation.

Introducing "Deep vectorised operators for pulsatile hemodynamics estimation in coronary arteries" (published in Computer Methods and Programs in Biomedicine).

YouTube video by Bastian Grossenbacher-Rieck Shapes, Spaces, Simplices, and Structure: Geometry, Topology, and Machine Learning

If you are excited about representation learning, point clouds, higher-order complexes, or just curious about geometry/topology, this talk is for you!

👉 youtube.com/watch?v=frjm...

🧵4/5

I am happy to announce that I have started a postdoc position at @aim-lab.bsky.social and @munichcenterml.bsky.social, Technical University of Munich under the supervision of @danielrueckert.bsky.social. Looking forward to this new chapter 🥨🎓

I also found out that William Kingdon Clifford (after whom Clifford algebra is named) is buried in London, so I visited his grave to pay respects.

...and to the organisers (including @simofoti.bsky.social, @valegiunca.bsky.social, @daniel-platt.bsky.social and @pragya-singh.bsky.social) for this great opportunity.

Three weeks ago I had the pleasure of mentoring a project at @logml.bsky.social. We looked into cortical surface parcellation from a geometric angle and had a lot of fun! Thanks to my project group (including @mustas.bsky.social and @zhengdiyu.bsky.social) for the trust...

Special thanks to my committee @erikjbekkers.bsky.social, Oscar Camara, Remco Duits, Bernard Geurts, Michel Versluis, and Jolanda Wentzel, my paranymphs @dieuwertjea.bsky.social and Patryk Rygiel and of course my supervisors @jelmerwolterink.bsky.social and @christophbrune.bsky.social!

It's already been two months but better late than never: I have successfully defended my PhD 🎉 It's been a roller coaster ride and I enjoyed every minute of it! I am forever grateful for the awesome people I met along the way (you know who you are ✌️).

🚨 New analysis: 257 suspicious X/Twitter accounts identified, spreading among others far-right narratives, anti-migration, anti-vaccine, and climate disinformation in the period before and after the 2025 German federal election.
👉 Report: dtecbw.de/sparta/
#Disinformation #btw25

Simple idea to filter out (some) LLM-written papers with non-existent references: check the reference file agains DBLP and desk reject if a certain % of citations doesn't exist.

I have cleaned up the notebooks for my course on Optimal Transport for Machine Learners and added links to the slides and lecture notes. github.com/gpeyre/ot4ml

Enschede crowd you don't want to miss this! Join us for two exciting talks by Jolanda Wentzel and @erikjbekkers.bsky.social. As a bonus, you get to see me in "appropriate traditional dress from [my] country of origin".

📅 28 May
📍 UTwente campus

We are looking for reviewers to serve on the Program Committee of the Symmetry and Geometry in Neural Representations (NeurReps) workshop at NeurIPS 2025! If you are interested in participating, please fill out this form: forms.gle/wgUXHT7Mq71n...

@neurreps.bsky.social

Equivariant neural networks can effectively model physical systems by naturally handling the underlying geometric quantities and preserving their symmetries, but scaling them to large geometric data remains challenging. Naive downsampling typically disrupts features’ transformation laws, limiting their applicability in large scale settings. In this work, we propose a scalable equivariant transformer that efficiently processes geometric data in a coarse-grained latent space while preserving E(3) symmetries of the problem. In particular, by building on the Geometric Algebra Transformer (GATr) and PerceiverIO architectures, our method learns equivariant latent tokens which allow us to decouple the processing complexity from the input data representation while maintaining global equivariance.

ICLR 2025 MLMP best poster award goes to "ViNE-GATr: scaling geometric algebra transformers with virtual nodes embeddings"! Congratulations @sukjulian.bsky.social, Thomas Hehn, @arashbehboodi.bsky.social, Gabriele Cesa!

Join one of the exciting projects led by: @ffabffrasca.bsky.social, Caterina Graziani, Anthea Monod & Omer Bobrowski, Guadalupe Gonzalez & Chirag Agarwal, @sukjulian.bsky.social, Vijay Prakash Dwivedi, Michael Murray, Yasha Ektefaie, Edward Hirst, Tancredi Schettini Gherardini

🤹 Excited to share Erwin: A Tree-based Hierarchical Transformer for Large-scale Physical Systems

joint work with @wellingmax.bsky.social and @jwvdm.bsky.social

preprint: arxiv.org/abs/2502.17019
code: github.com/maxxxzdn/erwin

Synthetic data to enhance new treatment uptake for acute ischemic stroke - Responsible Digital Transformations Multiple (pre-)clinical trials are required in the regulatory trajectory to introduce new medical devices in clinical practice. The process is widely acknowledged as sub-optimal, resource-heavy, and t...

🚀 Excited to be part of this project led by Praneeta Konduri! We’re using state-of-the-art generative models to create synthetic vasculature geometries—pushing stroke treatment development forward while cutting down on patient data reliance. Exciting stuff! 😃 rdt.uva.nl/research/res...

Detection of Epileptogenic Focal Cortical Dysplasia Using Graph Neural Networks This study evaluates the efficacy and interpretability of graph neural networks in automatically detecting focal cortical dysplasia lesions on magnetic resonance imaging scans.

Very pleased to officially introduce MELD Graph, a novel AI tool for the detection of subtle focal cortical dysplasia (FCD) lesions in epilepsy patients.

Check out our paper published in JAMA Neurology yesterday! 😀https://jamanetwork.com/journals/jamaneurology/fullarticle/2830410

What's a bit scary is the extent to which the right-wing extremist AfD's posts seem to dominate German-language socal media (in quantity, not quality). I really hope the right people see this and it is an eye-opener to them. We must resist, collectively.

Germans (or everyone interested in the upcoming federal election): this website does live monitoring of social media discourse (mostly X, formerly Twitter) by automatically labelling posts as friendly or hostile towards political parties and candidates. Super insightful!

ARD-Mittagsmagazin vom 24. Januar - hier anschauen +++ Nach Unfall mit Frachtschiff: Neues Schleusentor wird eingebaut +++ Kulinarisches aus aller Welt: 'Texas Barbecue' +++ Milde Winter ohne Schnee: Skigebiete schließen Liftanlagen +++

Im ARD-MiMa hat @jasminriedl.bsky.social heute spannende Einblicke in unser #ProjektSPARTA gegeben und darüber gesprochen inwiefern der Twitter/X-Algorithmus den Wahlkampf zur #btw25 beeinflussen kann.
🎥 Zum TV-Beitrag: bit.ly/ARD_MiMa
📊 Unsere Analysen: dtecbw.de/sparta/
#Bundestagswahl2025

The Developing Human Connectome Project: A Fast Deep Learning-based Pipeline for Neonatal Cortical Surface Reconstruction The Developing Human Connectome Project (dHCP) aims to explore developmental patterns of the human brain during the perinatal period. An automated processing pipeline has been developed to extract hig...

Need cortical surface extraction which is time efficient while being more anatomically correct than classic pipelines? Try Qiang Ma's pipeline - tuned for neonatal (dHCP) but we are already using it for fetal and works on ya HCP arxiv.org/abs/2405.08783 github.com/m-qiang/dhcp...

MCML Director Prof. Daniel Rückert has been awarded the Gottfried Wilhelm Leibniz Prize 2025!

The Prize is endowed with 2.5 million euros by the Deutsche Forschungsgemeinschaft (DFG) - German Research Foundation.

Congratulations!

@danielrueckert.bsky.social
📸 Juli Eberle

I created a starter pack of accounts to follow for AI in medicine.

go.bsky.app/TPXXbXp

Let me know if you would like to be added or have more suggestions 😊

I will be in Amsterdam doing an internship at Qualcomm AI Research until end of February. Hit me up if you want to link! 🚲🌷🇳🇱

Excited to see if the community can gain some traction on here 🤙