New paper out! Together with my postdoc advisors @esoubies.bsky.social and Cédric Févotte :) We study the optimization landscape of L0-Bregman relaxations, which naturally arise when one optimizes non-quadratic fidelity terms coupled with the L0-norm. 1/n

New on arXiv: Why noise matters in learned regularization! Data-driven regularizers fail without noise structure. 🎯 Tikhonov, Lavrentiev, quadratic — not equal under non-white noise. 🤫 Spoiler: Optimal reconstruction needs proper noise modeling. 👉 Regularizer choice = structure and noise.

What does it mean to do mathematics? Read my personal answer here: web.mathi.uni-heidelberg.de/content/uplo... #mathsky #math #question

So you agree that the sine function is not decreasing, but you won't say that it's non-decreasing? But alas math has such a precise language! 🧮

Ta daaa! I give you the Lorentz Attractor thanks to the free pattern from #ThreadedTheorems and also @samjshah.bsky.social whose posted work I admired. #StemBroidery #MathArt #MathSky

Join us for a panel on “Careers in Mathematics of Data Science” tomorrow morning!

Our latest WebMDS webinar was a blast! Check out our panel discussion of "Careers in Mathematics of Data Science" on youtube: youtu.be/qdlJJLT4ddQ?...

Register at siam.zoom.us/webinar/regi...

Save the Date! The SIAM Activity Group on Data Science (SIAG/DATA) invites you to join us for a panel on the topic: Careers in Mathematics of Data Science Tuesday, August 12, 2025 | 9:00 AM PT | Zoom Webinar Panelists: Marta D’Elia, Elizabeth Newman, David Uminsky

I've invented a simple, lovely math puzzle I expect every AI fails: Suppose you're a mathematical sailor at sea on a boat that has a perfectly cylindrical hole in the floor. All you brought is a collection of every p norm ball except p=2 (drat!). What do you do to cork the hole and save yourself?

A: A mathematical statement/concept makes intuitive sense. B: There is a sound framework where a mathematical statement/concept can be proven/defined. I tend to believe that A ⇒ B, but not vice versa. (Thinking about Dirac's delta and the Banach-Tarski paradox, for example.) #🧮

I still did not subscribe to d(Amazon)/dx…

Die Frage P=NP? ist eines der Milleniumsprobleme und somit lohnenswertes Ziel für Hobby- und Profimathematiker:innen. www.claymath.org/millennium/p... 🧵

I sample points uniformly on the sphere and then look at them on a map (in longitude and z-coordinate). STILL UNIFORMLY DISTRIBUTED! 🤯 (Quite a nice fact that is special to three dimensions: Each coordinate of a random vector uniformly on the uni sphere is uniform in [-1,1].) #🧮

The recording of the @siagdata.bsky.social webinar "Teaching the mathematics of data science" is online! youtu.be/nynELwnA8ow?...

On Parameterizing Optimal Transport with Elastic Costs with Marco Cuturi youtu.be/tltdW8L68PI

The Data and Science of Elections with Moon Duchin youtu.be/ykk4ddjukoA

Smartphone Privacy: How to Learn from Distributed, Private Data with Jelani Nelson youtu.be/YduytyEvTQA

The Emerging Science of Benchmarks with Moritz Hardt youtu.be/ySXQm7NIVN8

Plenary videos from SIAM MDS 2024 just dropped! Check out these great talks:

👇 the logic of capitalism

\addbibressource{tand_gopt2074_1247.bib} \addbibressource{S0024379525000242.bib} …

Recently learned from @gro-tsen.bsky.social that Claude Shannon got the (Alfred) Noble prize for his master's thesis! (He did not get the Nobel prize though…) en.wikipedia.org/wiki/Claude_...

Start saying "conclave" instead of "using the axiom of choice" when constructing large cardinals

Oh, and btw, if you know a resource where I can find a proof of the existence of the SVD of compact operators that does not use the spectral theorem for selfadjoint compact operators, I would like to know! If you know a proof that avoids using eigenvectors, I would be even more curious to know!

After almost six years I wrote another blog post! regularize.wordpress.com/2025/05/06/p...

Reminder: first WebMDS event Tuesday, May 6 at 9 am PT! Join us for a panel on “Teaching the Mathematics of Data Science”!

That problem is smooth. And if it's not, it is differentiable everywhere. And if it's not, we avoid the kinks almost surely. And if we don't, what is computed is a subgradient. And if it's not, it approximates one. And if that's not true, who cares? The loss went down.

Theorem: there is a theorem whose proof is shorter than its own statement. Proof: I could have said "this one" and stop there, but just for fun, I will refer to the next one instead. Theorem: there is a theorem that says nothing and still has a proof. Proof: this one.

Panelists: Nathan Alexander, Mine Çetinkaya-Rundel, Tammy Kolda, Rachel Levy This is a free, one-hour virtual panel open to all. Advance registration is required. Stay tuned for the registration link!

Save the Date! SIAG/DATA is thrilled to announce the launch of our new webinar series: Webinar on the Mathematics of Data Science (WebMDS)! Join us for the inaugural event – a panel on: Teaching the Mathematics of Data Science Tuesday, May 6, 2025 | 9:00 AM PT | Zoom Webinar

Energyachusetts = Massachusetts times Speedoflightachasetts²

o1 musing to itself in German: "Sister, I am working on considering all possibilities."

SIAM has announced the 2025 fellows and we are happy that among them there are the six member of SIAG/DATA David Gleich, Lior Horesh, Youssef Marzouk, Jonathan Mattingly, Stefan M. Wild, Mohammed J. Zaki Congratulations! www.siam.org/publications...

Kenne ich aus Mathe wirklich nur so. Bei mir gibt's Bonuspunkte für alle, die Fehler in meinen Skripten finden - aber nur, wenn Sie diese ins gemeinsame Forum der Vorlesung schreiben, damit ihn auch alle sehen und nicht selbst finden müssen.

TFW when you find out that the copy editor has confirmed with a thesaurus that "learning" and "understanding" are synonyms and has edited your 50 page text about machine learning accordingly

Something about the notion of smoothness is cursed. What is the smoothness of f(x) = |x|ᵃ? Well, it should be (and is) kind of a-Hölder - except when a = 2,4,6,8… when it's C^∞. #🧮

Rebranding hand waving as vibe proving. #🧮 #mathsky

#mathtwitter please brag about your office space/best office related purchase/ your favourite quirky thing in your office- help a girl build a faux-Pinterest board

When making a decision, we often try to assign some value to a thing. Do you think that value is an actual value or somehow based on the derivative (i.e. change) of some valuable quantity? Example: Need to decide to go to Coffee shop A or B. 1/2

A friend included this anti-acknowledgement section on her PhD thesis. She also added the proposition: “Systematic bullying and undermining of girls and women in STEM starts early on and is the reason why they do not stay in science and related fields.” Absurd we still need to go through this 🧪👩‍🔬

👋 The SIAM Activity Group Data Science is here! The group is run by the officers (row wise top to bottom) Jamie Haddock (secretary) Dirk Lorenz (vice chair) Rebecca Morrison (program director) Lars Ruthotto (chair) (www.siam.org/get-involved...)

That's exactly how I motivate image compression in my image processing lecture. Why should a picture take up megabytes of data if it's worth only a thousand words? JPEG2000 achieves just that (while JPEG usually comes down to tens of thousands of words).

Hi there! 👋 We're #HerMathsStory + want to empower women through stories & blog entries to pursue studies and careers in maths, spark their curiosity + transmit enthusiasm for technical subjects on an informative, non-judgemental, honest, motivating + enjoyable platform. #introduction #womeninmaths

Hey bskys - does anyone know about progress on formalization of applied fields of math like numerics and optimization? (@damekdavis.bsky.social started with gradient descent, but I haven't seen anything else in that direction…) #mathsky

It's the International Day of Women and Girls in Science. How are you marking the day? (#WomenInSTEM #STEM #InternationalDayofWomenAndGirlsInScience)

happy international women & girls in science day :)