Fast and Easy Infinitely Wide Networks with Neural Tangents (opens in new tab)

(ai.googleblog.com)

15 pointsrainboiboi6y ago3 comments

3 comments

3 comments · 2 top-level

dgreensp6y ago· 1 in thread

Can someone explain the implications of this for performance, or what we can do now that we couldn’t do before?

Well, they seem to be claiming it allows insight or understanding into how the networks work. However, they don't seem to demonstrate that as such.

rllin6y ago

It really looks more and more like JAX is the internal winner after the tf 2.0 fiasco

j / k navigate · click thread line to collapse