Umberto Picchini

Full Professor in Mathematical Statistics @StatsChalmersGU | simulation-based inference | Bayes | stochastic dynamical systems | @uPicchini.bsky.social

[Umberto Picchini @uPicchini]

Cool!

[Jeremias Knoblauch @LauchLab]

📢 Post-Bayesian online seminar series coming!📢

To stay posted, sign up at
https://t.co/Tx3ecXKdC8

We'll discuss cutting-edge methods for posteriors that no longer rely on Bayes Theorem.
(e.g., PAC-Bayes, generalised Bayes, Martingale posteriors, ...)

Pls circulate widely!

[Umberto Picchini @uPicchini]

for those migrating to Bluesky: these days me, as well as many others, are getting dozens of new Bluesky followers every day. But please before following anyone, write something about yourself in your profile, and have a profile pic. You are more likely to get followed-back

[Umberto Picchini @uPicchini]

In Göteborg you can choose what services to use to enhance your predictive power #realpic

[Adam Kucharski @adamjkucharski]

My new post on the place to increasingly be (and not to be) if you're interested in science, data, research, health, policy...

https://t.co/iaaFSaHLF5

[Approximate Bayes Seminar @approxbayes]

A reminder of our talk tomorrow by Ullrich Köthe (University of Heidelberg) on "Free-form flows for physics-informed generative modelling". Sign up here https://t.co/QIcmNUXjj3 to get the link to join.

[Umberto Picchini @uPicchini]

“Either you perform and bring profits to the company, or it’s more convenient to leave.”

[Umberto Picchini @uPicchini]

regarding the publisher MDPI: a tragedy put MDPI into the spotlight
"Performance metrics for editorial staffers prioritize the quantity of published manuscripts over their quality [...] It incentivizes us to push towards publication instead of rejection" https://t.co/Q1zI4BVDx4

[Umberto Picchini @uPicchini]

Please follow the new X account of the Approximate Bayesian Inference seminar series, to learn of advances in simulation-based inference.

[Richard Everitt @bayesian_stats]

The One World Approximate Bayesian Inference (ABI) Seminar is on X. https://t.co/q6R9Me3LRz

[Umberto Picchini @uPicchini]

recent *review paper* on (neural) amortised simulation-based inference.
I haven't seen it publicized much here among my connections, so there you go

[Andrew Zammit-Mangion @andrewzm]

Had a great time writing this review paper with former student Matthew Sainsbury-Dale and @HuserRaphael on "Neural Methods for Amortised Parameter Inference". Hope it's found as a useful resource, I definitely learned a lot writing it! #rstats https://t.co/o0sfQ5iisQ

[Umberto Picchini @uPicchini]

it generated a 15 minutes podcast involving a realistic discussion between two persons about the paper. It was an engaging conversation, with very realistic voices, interruptions, accents. Incredible. 4/n

[Umberto Picchini @uPicchini]

yesterday I refereed a paper, read it from start to finish as usual, and sent the review to a journal. This morning I read on LinkedIn of the NotebookLM project by Google, an AI "research assistant". You can upload a paper and "interrogate" it. 1/n

https://t.co/78lridPoAY

[Umberto Picchini @uPicchini]

When the AI Notebook returns comments, it also cites the exact paragraphs in the paper that are relevant for the comments.

You can see where this is going. You can also use it to extract a summary of a paper you wish to read, and "scaringly enough"....3/n

[Umberto Picchini @uPicchini]

tweeting this so I can also find it again.

Pen and Paper exercises in ML *WITH SOLUTIONS*

By Michael Gutmann

https://t.co/lPucR3Iuw9

[Umberto Picchini @uPicchini]

Harold Jeffreys introduced the | notation for conditional probabilities.

Learned from https://t.co/regR1oDq0j

[Umberto Picchini @uPicchini]

@tdietterich yeah it's a tough call. The previous attempt I am aware was by @researchersone

[Umberto Picchini @uPicchini]

Brad Efron's 13 rules for giving a really bad talk

(from https://t.co/r7g1lVxROY)

Thanks to Robert Grant for finding this gem

[Umberto Picchini @uPicchini]

a 129 pages review paper on optimal experimental design (thanks @FForbes70 for the pointer!)

[Stat.CO Papers @StatCOupdates]

Xun Huan, Jayanth Jagalur, Youssef Marzouk. [https://t.co/IhnC1wHNMM]. Optimal experimental design: Formulations and computations. https://t.co/joeDj0qTbn

[Umberto Picchini @uPicchini]

@MarvinSchmittML Perhaps you could give your take on alphaXiv, if you see it as something potentially big coming, since it is backed up by Stanford.
This is a topic falling more broadly into "academic life" than data science, i know ....

[Umberto Picchini @uPicchini]

nice package for visualising of the outputs of SMC, IS and more, by @bayesian_stats

For example, the pics shows the genealogy of 25 resampled particles

https://t.co/n20NAlHSSC