For small step sizes, mirror descent is equivalent to a gradient descent on a Hessian manifold.

If I understood 65% of this instead of 35% of this, I think it'd save ~90hr of my computer running at full tilt in my current research

Does mirror descent correspond to a discretization of this flow ?

Yes, a very smart one actually (for instance it preserves all the thm of speed of convergence as usual gradient descent)

How do you make this kind of animation? I mean by which software?

fucking cool! better than any pussies I have got.

@readwise save thread @SaveToNotion #thread @threadreaderapp unroll @memdotai mem it

@gabrielpeyre @readwise @SaveToNotion @memdotai @CeoSpaceY Hi! please find the unroll here: Talk to you soon. 🤖