News for hackers. Stuff that matters.
Re:Curse of dimensionality by Brashman on Wednesday May 30, 2012 @08:24PM (#4045624)
One example is suppose I'm trying to tune some parameters and I'm going to do this by sampling. If I have only one parameter which is between 0 and 1 to tune, then I can find the best choice to within .01 by trying 100 values (each .01 increment).

Now instead, consider what happens when I have 100 parameters (high dimensionality). Now with 100 samples, I can't really learn much. Even if I wanted to try only 2 values in dimension (say 0 and 1), there would be 2^100 (roughly 10^30) different possibilities to try.

I would say the key line from Wikipedia is that, "when the dimensionality increases, the volume of the space increases so fast that the available data becomes sparse."

There are no comments on this comment yet.