Differentiable programming in Kotlin

So, Facebook started an effort late last year to achieve a differentiable programming system in Kotlin. I am not quite sure how they are going to achieve nice interoperability with PyTorch, given that Kotlin is a JVM platform (or do they even want to achieve this kind of interoperability).

Of course, this is also a bit late in the game, with full-scale differentiable programming available in Julia in 2017 and in JAX/Python in 2018.

But it is even more funny that a task to create a differentiable programming system for Kotlin is already done by a very cool kid, who've done it single-handedly for his Master Thesis! (He has also rejected a job offer from Google, just doing PhD in Canada instead.) He is also writing some very interesting blog posts, about differentiable programming and other things. (Links in the comments.)

This world is something :-)

P.S. People sometimes call this differential programming and sometime they call this differentiable programming - I am not sure, do they settle on "differentiable" at the end of the day?

Лопать взрывчатку, чтобы похудеть и согреться!

Скотт Александр написал прелестную статью про историю изумительно вредной гадости (2,4-Динитрофенол), которая вполне успешно работает в качестве таблетки для похудения ("если выживешь, то похудеешь, но, может быть, ещё ослепнешь"). Удивительно, что правительства так и не смогли её вполне запретить, и особо одарённый народ продолжает её жрать mostly for bodybuilding purposes и травиться.

Теперь, разумеется, всякий biotech пытается сделать более безопасную версию этой штуки.

"It next shows up in history books on the Eastern Front of World War 2, where Soviet soldiers would - I can’t believe I’m writing this - take it to keep warm. There’s something quintessentially Russian about this, like a cross between the Platonic essences of AK-47s and Krokodil."

Ссылки и другие яркие цитаты - в комментариях.