iiinova — where your digital self keeps evolving

Reposted by Compute! Paris

🇫🇷 Jupyter is heading to Paris! Jupyter be front and center at Compute! Paris 2026 (Nov 25–26, Sorbonne Université), with an opening keynote by Fernando Pérez and a dedicated "Planet Jupyter" track. blog.jupyter.org/posts/2026/j... @computeparis.bsky.social @fernandoperez.org

compute.events

Compute! Paris 2026

Understanding your tools and algorithms is what gives you agency in a digital world. Paris, November 25–26, 2026.

064

Compute! Paris @computeparis.bsky.social · 18h

Every finding, all software behind, even the research field's FAIR rules, all get published similarly, turning independent claims into a connected, reusable graph of evidence, with verified citations. Perfect for giving humans and AI agents solid scientific grounding. Get your ticket!

compute.events

Compute! Paris 2026

Understanding your tools and algorithms is what gives you agency in a digital world. Paris, November 25–26, 2026.

000

Compute! Paris @computeparis.bsky.social · 18h

The talk draws on a year spent replicating 30+ published claims in ecology, biodiversity, and climate science. Recurring failure modes surfaced (biased representation choices, priors that strongly sway results, hidden train/test leakage), each catalogued with how to catch it.

platform.sciencelive4all.org

Science Live Platform

100

Compute! Paris @computeparis.bsky.social · 18h

The smallest unit of science to trust isn't a paper, it's a single claim. At Compute! Paris, Anne Fouilloux and Jean Iaquinta explore "nanopublications": small, cryptographically verifiable, machine-readable claims you can cite, link, and trust.

100

Compute! Paris @computeparis.bsky.social · 02/10/2026

Welcome to our new sponsor: @probabl.ai www.probabl.ai

000

Compute! Paris @computeparis.bsky.social · 02/10/2026

A tale of taming the cascading effects of one seemingly innocent design choice, classy engineering all the way through. Get your ticket and join the fun! 📍 Compute! Paris 2026 — Nov 25–26 huggingface.co/blog/parquet... 👉 compute.events/paris2026

compute.events

Compute! Paris 2026

Understanding your tools and algorithms is what gives you agency in a digital world. Paris, November 25–26, 2026.

000

Compute! Paris @computeparis.bsky.social · 02/10/2026

Rows untouched by an edit fall back into the exact same pages as before, remain byte-identical, and deduplicable nicely. Simple examples first: append, insert, delete, and the actual bytes saved on each. Then the real numbers, straight from that 100PB+ platform.

100

Compute! Paris @computeparis.bsky.social · 02/10/2026

At Compute! Paris, Quentin Lhoest, maintainer of the popular datasets library, introduces Content Defined Chunking to Parquet, ending a page based on the row data itself rather than counting rows. This makes page boundaries, and thus Parquet file byte content, more robust to change.

100

Compute! Paris @computeparis.bsky.social · 02/10/2026

This in turn breaks deduplication: storage systems that deduplicate, crucial for running 100PB+ stores like Hugging Face's, work by spotting identical bytes and storing them only once, so changing all bytes means nothing is left to deduplicate.

huggingface.co

Parquet Content-Defined Chunking

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

110

Compute! Paris @computeparis.bsky.social · 02/10/2026

One row inserted or deleted seems innocent enough, but with ordinary Parquet, it shifts every row after it into a different page than before. And once a page's rows change, its compressed bytes come out entirely different.

110

Compute! Paris @computeparis.bsky.social · 30/09/2026

This talk makes the case that an open GEE alternative isn't out of reach anymore, and that everyone can join in making it a reality. Get your ticket! 📍 Compute! Paris 2026 — Nov 25–26 jupytergis.readthedocs.io/en/latest/ 👉 compute.events/paris2026

compute.events

Compute! Paris 2026

Understanding your tools and algorithms is what gives you agency in a digital world. Paris, November 25–26, 2026.

000

Compute! Paris @computeparis.bsky.social · 30/09/2026

Deployable in-browser via JupyterLite and notebook.link ! The talk includes a real-world case study from CNES (French space agency): snow detection over the Pyrenees using Sentinel-2 data, comparing OpenEO’s reproducible processing graphs with Xarray-driven computation with jupyter-tiler.

notebook.link

Notebook.link - Instant Computing, Anywhere

No setup, no barriers. Run, share, and collaborate on Jupyter notebooks directly in the browser.

110

Compute! Paris @computeparis.bsky.social · 30/09/2026

At Compute! Paris, Guillaume Eynard-Bontemps, Vincent Sarago & Martin Renou (GeoJupyter community) present an open alternative from pieces already maturing across the Pangeo, CNG, and OGC ecosystems: JupyterGIS (frontend), TiTiler (backend), Xarray, STAC, and OpenEO (under the hood).

jupytergis.readthedocs.io

JupyterGIS — JupyterGIS

JupyterGIS

111

Compute! Paris @computeparis.bsky.social · 30/09/2026

For years, Google Earth Engine has been the only real option for scientists who want to run algorithms across petabytes of satellite data and see results fast. That's starting to change.

132

Compute! Paris @computeparis.bsky.social · 29/09/2026

Welcome to our new sponsor: Inria www.inria.fr/en

inria.fr

Accueil | Inria

000

Compute! Paris @computeparis.bsky.social · 28/09/2026

We are looking forward to a solid tool to build with, and a nice story to hear. Get your ticket! 📍 Compute! Paris 2026 — Nov 25–26 github.com/FBruzzesi/po... 👉 compute.events/paris2026

compute.events

Compute! Paris 2026

Understanding your tools and algorithms is what gives you agency in a digital world. Paris, November 25–26, 2026.

000

Compute! Paris @computeparis.bsky.social · 28/09/2026

While that's all neat, the talk's spicy part is the design history: abstractions they built and removed, parameterizations they flipped, the null-vs-panic contract they rewrote after hitting bad rows in production at scale. Benchmarks against scipy + Polars included.

100

Compute! Paris @computeparis.bsky.social · 28/09/2026

At Compute! Paris, Francesco Bruzzesi introduces polars-stats, a Polars expression plugin doing exactly that, Rust (statrs) underneath, column-valued parameters as a first-class feature, sampling reproducible across platforms.

github.com

GitHub - FBruzzesi/polars-stats: Polars expression plugin that exposes `scipy.stats`-style probability distributions

Polars expression plugin that exposes `scipy.stats`-style probability distributions - FBruzzesi/polars-stats

101

Compute! Paris @computeparis.bsky.social · 28/09/2026

Statistical distributions directly in Polars. No .to_numpy(), no round-trip through scipy.stats, no broken laziness or GIL lock. Just Normal(mu=..., sigma=...).cdf(x), one distribution per row, fully vectorized, lazy all the way through.

100

Compute! Paris @computeparis.bsky.social · 25/09/2026

You'll understand the Grammar of Graphics underneath (marks, encodings, the four encoding types), how to keep a chart responsive on gigabytes of data by pushing aggregation upstream to Polars or DuckDB, and leave with a sharper eye for every chart you see afterwards. No Altair experience required.

110

Compute! Paris @computeparis.bsky.social · 25/09/2026

Every chart has a purpose, every choice gets a moment: which visual channel to use, whether the scale is honest, whether an interaction earned its place, all worked through live with Altair.

100

Compute! Paris @computeparis.bsky.social · 25/09/2026

Slow down! Drive-by-plotting is not allowed at Compute! Paris this November: @amy8492.bsky.social takes you through deliberately visualizing Open Food Facts, a real, messy, multi-gigabyte open dataset.

110

Compute! Paris @computeparis.bsky.social · 23/09/2026

Get a ticket, join us in November! 📍 Compute! Paris 2026 — Nov 25–26 github.com/timweiland/L... 👉 compute.events/paris2026

compute.events

Compute! Paris 2026

Understanding your tools and algorithms is what gives you agency in a digital world. Paris, November 25–26, 2026.

000

Compute! Paris @computeparis.bsky.social · 23/09/2026

The nice part: Latte.jl lets you write your model once and pick INLA, TMB-style Laplace, embedded-Laplace HMC, or fall back to MCMC, a spectrum from fast to faithful unified through one interface. Tractable posteriors, be there, ditch the MCMC chains and compute what you couldn't before.

100

Compute! Paris @computeparis.bsky.social · 23/09/2026

Heĺl presnet Latte.jl: a probabilistic programming language that exploits the structure of latent Gaussian models, covering plenty of real-world use cases, to deliver posteriors orders of magnitude faster than what general-purpose sampling (MCMC) can do.

github.com

GitHub - timweiland/Latte.jl: Probabilistic programming for latent Gaussian models in Julia. INLA, TMB, and HMC-Laplace, all in one.

Probabilistic programming for latent Gaussian models in Julia. INLA, TMB, and HMC-Laplace, all in one. - timweiland/Latte.jl

100

Compute! Paris @computeparis.bsky.social · 23/09/2026

Stop Bruteforcing your Posteriors! Exploit known analytical structure! That's the title and pitch of Tim Weiland's talk at Compute! Paris this November.

111

Compute! Paris @computeparis.bsky.social · 21/09/2026

Spoiler: we will back out with a one-line workaround but what does this mean for computational reproducibility? Get your ticket! 📍 Compute! Paris 2026 — Nov 25–26 👉 compute.events/paris2026

compute.events

Compute! Paris 2026

Understanding your tools and algorithms is what gives you agency in a digital world. Paris, November 25–26, 2026.

000

Compute! Paris @computeparis.bsky.social · 21/09/2026

This November, at Compute! Paris, Christof will safely take us down the rabbit hole most of us would avoid alone: from discovery on a real HPC cluster in two standard genomics workflows (Seurat and Bioconductor), all the way to CPU-specific instruction sets quietly reshaping the math.

100

Compute! Paris @computeparis.bsky.social · 21/09/2026

Same seed, same code, same dependencies, even containerization, and still a different result. That's what Christof Seiler hit when running approximate PCA on single-cell genomics data: not cosmetics but numbers that shouldn't move, population counts shifting, rare cell types vanishing.

100

Compute! Paris @computeparis.bsky.social · 18/09/2026

At Compute! Paris, Arthur Andres lays out what makes a good rewrite candidate, walks the process from maturin setup to PyPI, and gives an honest look at where Rust wins big vs where friction eats the gains. The talk is illustrated with two real open-source libraries. No Rust experience required.

100

Compute! Paris @computeparis.bsky.social · 18/09/2026

Mature bindings with PyO3 and maturin, a zero-copy data interface via Arrow, and LLMs helping to write code. Never was it easier to (re)write Python library internals in Rust.

110

Compute! Paris @computeparis.bsky.social · 16/09/2026

And can TabICL tell "I'm uncertain because the world is noisy" apart from "I'm uncertain because I haven't seen enough"? Think about it and get a ticket to find out! 📍 Compute! Paris 2026 — Nov 25–26 github.com/soda-inria/t... 👉 compute.events/paris2026

compute.events

Compute! Paris 2026

Understanding your tools and algorithms is what gives you agency in a digital world. Paris, November 25–26, 2026.

000

Compute! Paris @computeparis.bsky.social · 16/09/2026

At Compute! Paris, @ogrisel.bsky.social, a driving force behind scikit-learn since the early days, puts this to the test against scikit-learn baselines: simplicity, compute cost, and statistical performance across sample sizes, scored with rules that cover calibration and sharpness.

github.com

GitHub - soda-inria/tabicl: TabICLv2: An open tabular foundation model

TabICLv2: An open tabular foundation model. Contribute to soda-inria/tabicl development by creating an account on GitHub.

100

Compute! Paris @computeparis.bsky.social · 16/09/2026

TabICL, a Tabular Foundation Model, pretrained once on millions of synthetic data-generating processes, automatically recognizes what shaped your data, and predicts accordingly. Out-of-the-box, well-calibrated predictions, classification & regression, that's the promise.

112

Compute! Paris @computeparis.bsky.social · 15/09/2026

Demos include "xtralab", a modern UI remix of Jupyter for AI use, and agentic workflows running entirely in the browser via JupyterLite. Many are wondering what scientific discovery and tooling will look like in the agentic era. Come hear what Jeremy has to propose!

100

Compute! Paris @computeparis.bsky.social · 15/09/2026

Way beyond simple chatting, they can inspect, drive, and adapt the UI live, running commands and executing code in backends or right in the browser. jtp.io (from quantstack.bsky.social ), Jupyter core maintainer, author of JupyterLite, and longtime community member, will show what that unlocks.

jtp.io

Home | Jeremy Tuloup

Welcome to Jeremy Tuloup's personal blog and website

110

Compute! Paris @computeparis.bsky.social · 15/09/2026

Everything in JupyterLab is a plugin, every action is a command, and kernels stand ready to run arbitrary code on demand. What’s great for humans turns out to be amazing for AI agents.

100

Compute! Paris @computeparis.bsky.social · 14/09/2026

With futurize(), you can pipe your code and parallelize it as-is, no rewrites needed. The same script runs seamlessly on a laptop, a cluster, or the cloud. 📍 Compute! Paris 2026 — Nov 25–26 futurize.futureverse.org 👉 compute.events/paris2026%22

futurize.futureverse.org

Parallelize Common Functions via One Magic Function

The futurize() function <doi:10.48550/arXiv.2601.17578> turns sequential map-reduce functions such as base::lapply(), purrr::map(), foreach::foreach() %do% { ... } into concurrent alternatives,…

100

Compute! Paris @computeparis.bsky.social · 14/09/2026

Meet "future", one of the top 0.6% of R packages and the go-to solution for scaling R with parallel tasks, from laptops to HPC. At Compute! Paris, Henrik Bengtsson (its creator and maintainer since 2015) will show how to simplify its API using R’s metaprogramming and transpilation.

111

Compute! Paris @computeparis.bsky.social · 09/09/2026

At Compute!, Marie Sacksick shows how to invert that dynamic: skills cut prompts, abstractions trim reviews, and the art of asking good questions, stays yours. Looking forward to seeing you in November. 📍 Compute! Paris 2026 — Nov 25–26 👉 compute.events/paris2026

compute.events

Compute! Paris 2026

Understanding your tools and algorithms is what gives you agency in a digital world. Paris, November 25–26, 2026.

000

Compute! Paris @computeparis.bsky.social · 09/09/2026

AI was supposed to do the laundry so we could do the art. In data science, that art is modeling, exploration and understanding the problem. Instead, too often agents get the fun, and we're stuck with the chores: reviewing code, firefighting it in prod.

100

Compute! Paris @computeparis.bsky.social · 07/09/2026

A live demo grades a database assignment start to finish, review and LMS collection included. Get your ticket, teach more with less. 📍 Compute! Paris 2026 — Nov 25–26 👉 compute.events/paris2026

compute.events

Compute! Paris 2026

Understanding your tools and algorithms is what gives you agency in a digital world. Paris, November 25–26, 2026.

000

Compute! Paris @computeparis.bsky.social · 07/09/2026

A. T. Darian from @quantstack.bsky.social built a version where: code runs sandboxed in the evaluator's own browser, student or teacher, the notebook file itself carries the grading criteria, state, and scores, and data is pulled and pushed directly from an attached LMS.

101

Compute! Paris @computeparis.bsky.social · 07/09/2026

Autograding notebooks usually means a server running student code, plus a database tracking submissions and scores. But what if you could skip both?

111

Compute! Paris @computeparis.bsky.social · 04/09/2026

GLIDE, an open-source Python package, uses Prediction-Powered Inference to scale ~100 human labels plus what the agent said into a number a statistician can sign off on. Live demo included, questions appreciated. Get your ticket! 📍 Compute! Paris 2026 — Nov 25–26 github.com/EmertonData/...

github.com

GitHub - EmertonData/glide: Helping humans ride the GenAI evaluation wave

Helping humans ride the GenAI evaluation wave. Contribute to EmertonData/glide development by creating an account on GitHub.

100

Compute! Paris @computeparis.bsky.social · 04/09/2026

Agents judging other agents are black boxes whitewashing black boxes. If you don't find this funny, come to Compute! and hear Gregoire Martinon present a more principled approach: injecting human understanding in a controlled way into the evaluation but still use agents to scale it up.

100