/sci/ - Science & Math

>>12532259
I'm sure we could list a bunch, say knot-theory-via-TQFT, or muh perfectoid spaces, or let's even say univalent types or ...Feynman path integrals to estimate the relearning distance in weight space of a deep neural networ. I'd say Hairers stochastic tools, but I don't know too much about it and I don't want all my examples to come from the same corner).
But they are all of the form that require a huge amount of background. As always, that which is deemed relevant will be cleared up and, made learnable and look not as obstroose in decades to come. Time progresses and new textbooks appear the proof old theorems via the new tools and suddenly they look fundamental. Say Noethers ideal theory for talking about numbers.
Computational complexity, Hilbert space theory or scheme theory or theory all looked extremely obstroose, once. Some ideas that didn't gain momentum from those times still look weird, because they were not developed.

That's my take for today.
Here's some plans for 2021:
https://youtu.be/UXTz-eFDmjc