Arash Behboodi

We are also actively recruiting for several positions, including for ML Engineers - take a look here:

careers.qualcomm.com/careers/job/...

Our internship hiring campaign has started! We have exciting opportunities at Qualcomm AI Research on various topics such as model efficiency, generative vision, agentic AI and reasoning, embodied AI, robustness and alignment.

Apply here:

careers.qualcomm.com/careers/job/...

The credit is really for @sukjulian.bsky.social, Thomas and Gabriele! Congrats to them!

This is a great achievement. Kakeya conjectures have been one of my favorite open problems. And I encountered it when digging deeper in GMT while reading the paper of Wu and Verdú on Rényi information dimension (and the relation of information theoretic and geometrical notions of dimension)

Talagrand Meets Talagrand: Upper and Lower Bounds on Expected Soft Maxima of Gaussian Processes with Finite Index Sets Analysis of extremal behavior of stochastic processes is a key ingredient in a wide variety of applications, including probability, statistical physics, theoretical computer science, and learning theo...

In this new paper with Yifeng Chu, we are asking what happens when you take a Gaussian process and heat it up a bit: arxiv.org/abs/2502.06709

Lean for Mathematicians 2025 Workshop Information Application Form

The "Lean for Mathematicians" workshop, aimed at training graduate students and postdocs in the use of the Lean proof assistant language for mathematics, runs June 16-27 2025 and is currently taking applications. sites.google.com/view/simonsl...

Definitely fascinating. There was a story about O1 that gave a compelling answer about Grassmannian packing problem but upon further investigation it turned out the answer was not correct (see Robert Heath post on Linkedin). This seems different though.

Initially it looked similar to o3-mini, spouting a whole lot of obviously irrelevant thoughts. Then it just stopped, so I assumed it had given up. But a few minutes later I returned to the page and found that it had given me a correct solution. At the end it wrote the following summary.

I've just tried out o3-mini-high on the following question.

I'm wondering if you can help me with the following maths problem. I'd like to find a property of pairs of distinct primes such that if X is any infinite set of primes, then one can find a pair of distinct primes in X ... 🧵

L'acquisition des fractions est une étape clé de l'éducation aux mathématiques.
Dans ce numéro du Passeur (toujours gratis!) la spécialiste Monica Neagoy explique comment rendre l'apprentissage des fractions facile et ludique.
t.co/x0UD8VzKbV
La vidéo est fantastique!
www.youtube.com/watch?v=SdZC...

#Science #Psychologie
« Les fondements cognitifs des #mathématiques : des intuitions universelles et précoces »
📌 Conférence d'Elizabeth Spelke (Harvard University)
⏰ Mercredi 22 janvier 2025, à 17 h 30
En direct 👉 lnkd.in/ehM3F2Tt

And one from Neri Merhav from a core information theory perspective: Statistical Physics and Information Theory

The other one I loved was Information, Physics and Computation by Mézard and @andrea-montanari.bsky.social , great book if you like to see the connection with BP and graphical models.

Indeed, this is a very nice piece. It is always interesting to see the nod to the relation of phyiscs and information theory. Let me mention some of the works I personally liked. First of course is Information Theory and Statistical Mechanics by Jaynes himself.

Very interesting. I hope such interactions grow between machine learning folks and those working on historical archives. I think some one like Lévi-Strauss or Barthes would have loved it.

Lorraine Daston is one the greatest minds in historical epistemology. Highly recommended to anyone interested in history of science.