Michal Rolínek

abstract math -- theoretical CS -- AI & ML research -- algorithmic trading. One of the folks behind #blackboxbackprop, now just a hobby researcher :)

[Michal Rolínek @MichalRolinek]

Bridging discrete optimization with deep learning. Maybe highly optimized (nondifferentiable) solvers for combinatorial problems will become great again 😍.

Differentiation of Blackbox Combinatorial Solvers https://t.co/YOzYOYhz20

[Akshay Agrawal @akshaykagrawal]

With @marimo_io , we're reinventing Python notebooks as reusable Python programs — execute as a script, deploy as a web app — instead of error-prone JSON scratchpads.

I've been meaning to write a blog post to explain more, but in the meantime: https://t.co/pNhr8kwRf4

[Marin Vlastelica 🤖🎸 @vlastelicap]

Glad to have given this talk in person at #ICML2023 and that many are interested in blackbox diff, it's gonna be online if you missed it, was a blast 💥!

[Jacob Buckman @jacobmbuckman]

@GaryMarcus Are you...serious? I posed almost exactly this argument last June, a day *before* I critiqued your work: https://t.co/IozYBm1r15

My critiques against your position still stand unchallenged:
https://t.co/YaYshVN9fV

[meowbooks @meowbooksj]

New template dropped @lunarflu1

[Michal Rolínek @MichalRolinek]

@CellTypist @AIMS_Rwanda @AIMS_Next Say hi to @AIMS_Rwanda !

[Michal Rolínek @MichalRolinek]

@Taibanganbakon1 @PoShenLoh Is this true?

[Michal Rolínek @MichalRolinek]

Final round of the Czech 🇨🇿 math olympiad is over. Out of 42 possible points, all 50 contestants received at least 7. The winner had 41. 🎉🎉

Chat-GPT scored a grand total of 0 points.

Congratulations to all human participants, you will be hard to replace by AI 😁🔥.

[Michal Rolínek @MichalRolinek]

@dpaleka Haha, good luck 😂.

[Michal Rolínek @MichalRolinek]

National round of Czech math olympiad is underway! Top 50 students vs. six hard problems 🔥🔥

Want to try one about sidelengths of an n-sided polygon?

(translation: prove the inequality! 😉)

[Lenka Zdeborova @zdeborova]

Kudos to Stephan Boettcher for debunking some of the recent claims of machine learning-based approaches outperforming traditional methods in hard optimization tasks. https://t.co/wdxWYM6lAe and https://t.co/cpLWzgHt6B I am also excited about ML, but let's do science, not PR!

[Zico Kolter @zicokolter]

Generative models and P vs. NP: A clickbaity thread 🧵

An important point that seems missing (as far as I've seen) in the debate over LLM "correctness" / "usefulness": generative models are most useful precisely in situations where creation is hard, but verification is easy. 1/N

[Georg Martius @GMartius]

My team was on TV 📺 yesterday evening ;-)
Really happy that the popular NANO series @3satnano at @3sat (D/A/CH TV channel) reports on our work on #robotlearning.

Thanks to the team and the @MPI_IS press office.

We are featured at ~21:15 in the clip:
https://t.co/9CkjKYzNbT

[John Nay @johnjnay]

French researchers converted their tax code into computer code

Compiles to Python & provides insights about "essence" of France's income tax computations

Government is officially transitioning to this for production.

Paper: https://t.co/vuzuxJ83XT
Code: https://t.co/x4oCryNVVl

[NEXTA @nexta_tv]

Retired General Petr Pavel is leading the elections in the Czech Republic. He is likely to become the country's new president.

Pavel is known for his pro-Ukrainian views, while his opponent, the scandalous oligarch Andrej Babiš, advocates cutting aid to Ukraine.

[Anselm Paulus @ICML 2026 @AnselmPaulus]

#Differentiating through discrete optimization solvers can be so easy!
They can often be treated as an Identity function on the backward pass. See our #ICLR2023 paper: https://t.co/hSXHb1pYob
Previous work, such as #Blackboxbackprop, requires an additional call to the solver.🧵

[Georg Martius @GMartius]

Wondering how to do exploration in continuous action spaces in #DeepReinforcementLearning?
We have an answer: Pink Noise Is All You Need ;-)
Checkout our ICLR2023 paper: https://t.co/DKJFbWB7ZG
and the following 🧵.

[Some theorems @CihanPostsThms]

[Hasse 1966]: Given x>0, let P(x) be the set of prime numbers ≤ x and
A(x) := {p∈P(x) : p divides 2ⁿ+1 for some n∈ℕ}.
Then lim_{x→∞} |A(x)| / |P(x)| = 17/24.

[Jacob Buckman @jacobmbuckman]

Are you a PhD student struggling to get a job or internship? Jealous of the success of your more-cited peers? More concerned with your career than doing good science? Here is a thread of eight invaluable techniques to "improve" your publication and citation metrics.

vv 🧵🧵🧵 vv

[Michal Rolínek @MichalRolinek]

@PMinervini Most definitely keep it simple with slurm.

[Georg Martius @GMartius]

Wow, our paper made it into the list of 3 finalists for the best paper award at @corl_conf. Congrats to the team @breadli428, @_sebastianblaes, @vlastelicap, Jonas, and Felix