The cat, again:
The cat, again:
The cat:
The cat:
Also, somebody convince Dan Goodman to get on Bluesky pls!
Also, somebody convince Dan Goodman to get on Bluesky pls!
Comment from Grimmer: "Gradient descent can accelerate (in big-O!) by just periodically taking longer steps. No momentum needed to beat O(1/T) in smooth convex opt!"
arxiv.org/abs/2309.09961
Comment from Grimmer: "Gradient descent can accelerate (in big-O!) by just periodically taking longer steps. No momentum needed to beat O(1/T) in smooth convex opt!"
arxiv.org/abs/2309.09961