Bernat Font

a paella

the best things are also thanks to fluid dynamics!

𝗖𝗼𝗺𝗲𝘁 𝗿𝘂𝗻𝘀 𝗰𝗶𝗿𝗰𝗹𝗲𝘀 𝗮𝗿𝗼𝘂𝗻𝗱 𝗝𝘂𝗽𝘆𝘁𝗲𝗿
Friends, I watched an AI agent complete my students' coding exercises in real-time, and I'm shook 🤯
I tested Perplexity's Comet browser on a Jupyter notebook with eigenvalue problems I assigned to my class… (video at the bottom of this thread)

109 de contingut, 143 total (index + refs + etc...)

Slides for the talk avaliable here!

b-fg.github.io/EFDC2_2025/m...

I'm loving this quarto + reveal.js for creating html presentations. Now that's portable!

Tomorrow at noon, I will be presenting at EFDC2 in Dublin! I will talk about our new mesh-free SINDy framework, collaboration with M. Gao and N. Kutz, for system dynamics identification on unstructured, scarce, and noisy data

🗓️ Session: tinyurl.com/c7up564z
🔗 Preprint: arxiv.org/abs/2505.16058

GitHub - adrhill/julia-ml-course: Julia Programming for Machine Learning course at TU Berlin Julia Programming for Machine Learning course at TU Berlin - adrhill/julia-ml-course

Just noticed my "Julia for Machine Learning" course got 2^8 stars on GitHub, yay! 🎉
github.com/adrhill/juli...

Reminds me that I need to start on the slides...

Exciting talks, brilliant minds, and a community that keeps pushing the boundaries of what’s possible with Julia.#JuliaCon2025 #JuliaLang #OpenSource #ScientificComputing #DayOne #CommunityVibes #TechConference #JuliaCommunity

CFD simulation of a whale tail

The WaterLily.jl article is now available on Computer Physics Comms!

@gabrielweymouth.bsky.social and I review our CFD solver's design, performance, accuracy, and also show some very cool examples (like the one below). Check it out!

🔗 doi.org/10.1016/j.cpc.2025.109748

#CFD #GPU #HPC #JuliaLang

Want to see what the flow past a semi-infinite array of randomly shifted and scaled circles looks like? I did too!

Code is modified from the examples here: github.com/WaterLily-jl...
#JuliaLang

FastFlows focuses on accelerating scale-resolving turbulent flow simulations combining #CFD, #HPC, and #ML. This enables the integration of such simulations into rapid industrial design cycles, providing accurate and physical solutions with reduced turnaround time.

I am happy to share that I have been awarded a Dutch Research Council (NWO) Veni personal grant!

www.nwo.nl/en/news/200-...

very cool!

Today’s vibecode achievement using @gabrielweymouth.bsky.social WaterLily. We went from 10 cylinders to 100 in no time at all; all on a standard laptop #vibecoding #julia #waterlily

Makie v0.24 focuses on a comprehensive overhaul of our internal architecture, specifically replacing Observables in our backend code with a more robust system. This fundamental change touches nearly every component of the codebase, making it one of our most ambitious refactoring efforts to date.

Makie v0.24 is out! One of the biggest architectural updates yet—bringing better performance, cleaner internals, and new plotting capabilities. A big step toward Makie 1.0. makie.org/website/blog...
#JuliaLang #OpenSource #Visualization #Plotting #HPC #MakiePlots #WGLMakie #Makie

How Bees, Beer Cans and Data Solve the Same Packing Problem (Gift Article) Trying to fit it all in? There’s a trick to it, even in 24 dimensions.

Dear #mathsky, if you repost this, it will let people see the article for free. Thanks!

www.nytimes.com/interactive/...

believe in*

I don't believe TOCs for slides, so that's fine by me. But I haven't given Typst a try yet, it's in my long To-do list..

making #quarto slides can be a bit painful, but the end result is amazing. Feels like #Beamer, but it comes with the benefits of #markdown and #html

LLMs with uncertainty quantification output with be nice too

ha, indeed! It was a gif.. nevermind.. thanks for the heads up!

v1.4.0 includes a new set of performance improvements and features such as
- Explicit LES subgrid-scale models
- Custom user-defined function that is called during every time step
- Easy to run an co-visualize with the new routine, viz!
- Custom dynamic boundary conditions and body forces

2/2

Flow past the Julia logo

New WaterLily version is out 🌀!
github.com/WaterLily-jl...

#JuliaLang #CFD #GPU #LES

1/2

Leveraging neural networks and automatic differentiation, mesh-free SINDy demonstrates robust PDE discovery and computational efficiency for a variety of 1D and 2D systems.

Mesh-free SINDy framework and 1D Burgers' equation as example.

Our latest preprint on "Mesh-free sparse identification of nonlinear dynamics" is out! In collaboration with Mars Gao and Nathan Kutz, we explore a novel SINDy framework to uncover dynamical systems from scarce, noisy, and non-uniformly sampled data.
🔗 arxiv.org/abs/2505.16058
#SINDy #PDE #SciML #ML

Em va passar el mateix amb una altra marca.. 🤦

Yeap.. I only started to like Overleaf when they introduced that feature tbh. Otherwise we cannot rely on a service like that, and here is proof 😑

#Overleaf is down and some are freaking out about a deadline coming right up... The standard work-local & push-to-remote approach is clearly not as widespread as I thought!

A comparison of automatic differentiation paradigms between Python and Julia: - In Python, one chooses the autodiff framework first (PyTorch / JAX), then the appropriate scientific library - In Julia, one writes the scientific library first, then one tries to make it compatible with several autodiff frameworks (Enzyme, Zygote, etc)

How to make #autodiff user-friendly? What lies beyond the safety of Python-world? Why does it matter for scientific machine learning?
All this, and more, in our latest preprint with @adrhill.bsky.social! Spoiler alert: it describes the most useful software I ever wrote.
arxiv.org/abs/2505.05542