Hi. Currently, I'm developing a NN and got stuck with the question of whether there are any approaches to measure statistical modeling capabilities of my NN? There are quite a lot of ways for modeling the same distribution, but is there any approach to figure out which way is actually the best (more capable)?
Did you use Bayesian neural nets?
Actually, NN are universal function approximators, which in theory means that nets can model any given function (and even its derivative) to an arbitrary level of precision. But in real life you are limited by the data and available computational resources. That is why different transformations can give you different results because they make the NN's job easier or harder. So, in general, there are no actual unified methods for measuring capability.
Just drop us an email to ...
Show more 
Log in with Google