Sometimes I wish I could do a master's internship again, and this is one of those times...

And of course big Thanks to the Jury including Nicolas Thome, Maarten De Vos, Alain Rakotomamonjy, @rtavenar.bsky.social and as president Agnès Desolneux .

Le dernier Ragnarok est génial mais quand tu est un fan qui a acheté les vieux qui ont pas la même taille ça fait moche dans la bibliothèque et ça c'est triste 🥲

Ender's game

So because you don't want large batches you solve the whole OT problem! Love it.

It's a great discussion because we all have different interpretation of what this means and what the consequences are.

Myself, I have absolutely no doubt that scaling works. If you have all the videos in the world and are able to train a model that can recall and merge any of them, then for sure...

This is why I love bsky! Totally agree with you. An important question of science is why and how is it working. This is what we should look at and promote as a community.

FOMO is our careers nowadays.

OK we should probably talk directly ;).

And yes I already checked the .h file it is very simple, so for POT we need a cython wrapper and a way to allows wheels compilation with eigen without exploding (when compiled from sources) but this would be very nice.

That was exactly my question to @nbonneel.bsky.social 10 se ago! I would erally like to have larger scale solvers even for low dims. I know that in some physics applications (OT on particles) scaling was a problem

Also I was wondering how well it works in higher dimensions (up to few hundreds) when you said you don't use the Gaussian slicing? Errors wrt true solver as a function of d and k would be very interesting

Hello, I'd be interested to have a Python/numpy wrapper of this awesome solver in POT. Would you be interested in helping out?