Is there any existing avenues of research on independence of bivariate kernel density estimation?

Kind of long, and I'm running out of time to read the whole thing, so sorry if this question is buried in the text, but it seems like several estimation methods are presented. Could you go into some inference of how each method compares and when each method is appropriate?

When using EM to estimate a gaussian mixture model, you have to specify the number of gaussians in your model. Do you have any recommendations on how to do that?
(Thanks, Rasit for putting up a copy! Louka, sorry I wasn't clear - I don't have a login for accessing the reserves. But thank you for the idea.)

You should be able to reach the paper through a copy on my webpage now:
http://www.cs.ucsd.edu/~rtopalog/chapter2.pdf

Robin, I'm not sure if this will help, but you can use UCSD's web proxy in order to access web sites as if you were on a campus computer: http://resnet.ucsd.edu/body_webproxy.html

Rasit, could you post the full reference for your paper? I can't access the course reserves from home, and I'd like to see if I can read it before I go back up to UCSD.
Thanks!