Previous | Next --- Slide 27 of 52
Back to Lecture Thumbnails
raymondx

This reminds me of ML algorithms that use gradient descent to minimize a multivariable function which is sort of the opposite of what is said on this slide.

barath

Actually, there is also gradient ascent in ML where you try to maximize a concave function rather than trying to minimize a convex function (Reaching the top of the mountain vs reaching the bottom respectively).

ahhuang

Like what was said previously, this is a common method for objective function minimization in ML, although it's usually gradient descent.

dchen1

I've only really encountered this in ML, I had no idea this stuff was in graphics too. I've never thought about what kinds of things in graphics would need optimization, can anyone give me a quick example maybe?

keenan

There's really no difference between minimizing a convex function and maximizing a concave function: you're just putting a minus sign in front of the objective.

(On the other hand, maximizing a convex objective or minimizing a convex objective may not be as straightforward!)