[Databionic-ESOM-User] overfitting

[Tali Raveh-S. <tal...@gm...>]

Hi,

I would like to use ESOMs for an application in which I have ~1000 points
(that are ~250 dimensional). From the manual I understood that one should
use few thousands of neurons. Wouldn't that lead to overfitting? If not,
why so?

I read in some of Ultsch's papers that increasing the number of neurons
does not directly increase the number of free paprameters since there are
still the toplogy constraints, but is this true also when the ration of
data points to neurons is more than 1:4? why?

Thanks
Tali