FORTRAN NOT DEAD

OK, we all know very well that Scientific Computing runs on FORTRAN: many high-performance math libraries like BLAS, many many-decade old simulation libraries and models in all sorts of fields — weather, oceanography, quantum chemistry, molecular biology, pharmacology and what not. Legacy stuff, right?

WRONG

Food and Agriculture Organization (FAO) of the United Nations recruited KU Leuven to rewrite an industry-standard model for crop growth in response to water (or lack thereof) from Pascal to FORTRAN:
https://github.com/KUL-RSDA/AquaCrop

Luckily some folks want to rewrite it to Julia:
https://discourse.julialang.org/t/short-term-scientific-programmer-for-crop-growth-modelling/109251

Почему кто-то вообще запомнил и записал, что Иисус Христос как-то раз взял да превратил воду в вино? Понятно, что у них там не было сети "Красное и Белое", но тем не менее... Это же не самое важное достижение в его жизни, правильно? Лечил слепых? Ерунда. Ходил по воде? Фокусы. Воду в вино превратил?! Наконец-то что-то полезное! О чём думали эти люди? Нет, реально — что происходило у них в головах, что вино оказалось в одном ряду с лечением слепых и воскрешением полумёртвых?

https://www.youtube.com/watch?v=RUJFd-rEa0k&list=PLP8iPy9hna6Tp3QV4akXAd23_O5Vjm_e2
Discussion:
https://discourse.julialang.org/t/does-julia-create-a-1-5-language-problem/107984

Julia famously claims to solve the "two languages problem": you use one language (Python) for high-level prototyping or configuring the actual computation graph (TensorFlow) and another language (C/C++) for low-level performant heavy lifting (NumPy). With Julia you allegedly can do both things in the same language.

Indeed, there are Julia libraries that outperform industry-standard heavily optimized decades-polished C/C++ counterparts (some ODE solvers for instance). And yep, you can program GPGPUs (both NVidia and AMD these days) without switching to special CUDA language extensions and a pipeline.

BUT. The authors point out that "friendly Julia" and "performant Julia" are very different styles even if the same language. Hence, 1.5 languages.

The observation not particularly fresh, the notion of "mechanical sympathy" floats around for quite some time. So the issue haunts pretty much any language: high-performance C or C++ is pretty different from "naive" or straightforward one. It might be even much more pronounced for Rust: "async Rust" is distinct enough for many people to avoid it altogether.

I don't know of a good solution for this discrepancy. Though every generation of languages have its try. We'll see how Mojo fairs in this regard.


Also to note: there's a Gaussian Process model evaluated on every spin of a car cylinder on a custom hardware... 🤯

https://statmodeling.stat.columbia.edu/2024/02/15/uncertainty-in-games-how-to-get-that-balance-so-that-theres-a-motivation-to-play-well-but-you-can-still-have-a-chance-to-come-back-from-behind/

Andrew Gelman got to (board) game design. Something's coming! 😁

https://twitter.com/SchmidhuberAI/status/1745475698737938543?t=_X-WsN5TCusvNMCRHkIxgA&s=19

Надо тоже перекатываться в AI — к 60 как раз успею подкачаться... 😁

I like https://pyre-check.org — it does a type inference and type checking for Python, and supports user-defined analysis rules for security or other potential issues (https://pyre-check.org/docs/pysa-basics/). It's implemented in OCaml following nice classic Abstract Interpretation framework and architecture explicitly abstracting away abstract domains.

And it's employed for actual high-volume production analysis at Instagram: https://instagram-engineering.com/static-analysis-at-scale-an-instagram-story-8f498ab71a0c

https://statmodeling.stat.columbia.edu/2024/02/21/scientific-publishers-busily-thwarting-science-again/

Scientific publishers rock again! So much benefit for society!

Registration is open for the Workshop on Homotopy Type Theory and Univalent Foundations:
https://hott-uf.github.io/2024/
Leuven, Belgium, April 2–3, 2024
Online attendance is possible but registration is required anyway,

https://www.pm.inf.ethz.ch/research/verifythis.html

Pretty famous annual VerifyThis competition will be held at ETAPS April 6th and 7th, 2024.
Participants are welcome to register and attend, (some) travel grants are available.

https://www.youtube.com/watch?v=XLtxXkugd5w

Curious. Apparently the "Wolfram Physics Project" isn't really about physics, but about a mathematical formalism capable of describing a wide spectrum of possible (or impossible) "physics". And yet it has some fundamental and generic features that lead to potentially testable consequences in our physics. 😊

Twelve of the sixteen Federal topographic maps covering the river corridor had gross errors. Thirty-five of the ninety-three rapids were unmarked and two dangerous falls were not on the maps at all. Twelve portages were missing and six were marked in the wrong location.

as of 1995 on a single river in Canada. Leading to many drownings over the years.

I think topographic maps have improved immensely across the globe since that time...

Как называется предприниматель, который практикует недеяние?

Пиз-деццкая комната

https://causalml-book.org/

#free #book on #causalinference and CausalML #machinelearning with a lot of Google Colab Notebooks in Python and R using EconML and DoubleML packages.

https://statmodeling.stat.columbia.edu/2024/02/28/a-suggestion-on-how-to-improve-the-broader-impacts-statement-requirement-for-ai-ml-papers/

For ML/AI folks writing papers. Some ideas on improving your "broader societal impact" section.

Дамочка! Недостаточно просто лежать в постели как бревно. При этом ещё нужно спать!

https://dmkpress.com/catalog/computer/mathematics/978-5-93700-271-6/

Не знал, что на русском вообще есть книги про SciLab, а это уже второе издание, сильно переработанное и дополненное. По сути — учебник для ВУЗов, с задачами для самостоятельного решения. Едва ли расскажет что-то новое заядлым пользователям Matlab/Mathematica/Scientific Python, но как введение в предмет с нуля может быть вполне.

#book

Яндекс умеет в юмор! 🤣

https://bobkonf.de/2024/en/program.html

The upcoming (March, 15) BOB conference features a keynote by Andreas Rossberg on formal semantics and its use for specifying WebAssembly (obviously), a tutorial by Andres Löh on Staged Programming in Haskell (with Typed Template Haskell), on Lean by Joachim Breitner and David Thrane Christiansen, on K Framework by Jost Berthold and Georgy Lukyanov, and some other interesting talks.

No live broadcast, but they seem to publish recordings on the talks' pages after the conference.

https://ig.ft.com/ai-animals/

Animals are smarter and more sophisticated than you think. Especially large mammals like elephants and whales. They use elaborate signaling systems aka language.

Now groups of researchers hope to decode the semantics of certain animal languages at least partially, with the help of modern Large Language Models. But first they have to collect and clean huge datasets, which presents gigantic challenge, both logistically and technically.

That kinda shifts your perspective on the elephants in a zoo...

Для ещё более слоупоков, чем я, сообщаю, что Bartosz Ciechanowski ещё год назад выкатил великолепное (как обычно) детальное интерактивное объяснение, как "работают" велосипеды:
https://ciechanow.ski/bicycle/

Кое-какие нюансы для меня были совершенно неизвестны и удивительны (кстати, недавно в чате обсуждали тросы вместо спиц и как это работает), а Бартош даёт ссылки на ещё более детальные научные исследования и эксперименты (с видео на YouTube).