Talk:Radial basis function network

What is a prototype data stream?

In https://en.wikipedia.org/w/index.php?title=Radial_basis_function_network&oldid=678776568#Logistic_map there is the sentence

The basic properties of radial basis functions can be illustrated with a simple mathematical map, the logistic map, which maps the unit interval onto itself. It can be used to generate a convenient prototype data stream. The logistic map can be used to explore function approximation, time series prediction, and control theory. The map originated from the field of population dynamics and became the prototype for chaotic time series

I don't understand what is meant with "prototype data stream". What is it? Would somebody please elaborate? --MartinThoma (talk) 12:09, 20 January 2016 (UTC)

Determining center vectors by clustering

The article says in the "training" section that center vectors can be determined by k-means clustering. Why would this work better than choosing random center vectors? Intuitively, a good center vector is one that is located at a "bump" in the output. Since the $\mathbf {x} (t)$ may be drawn from any distribution, they may cluster anywhere, not just in regions where the $y(t)$ have elevated values.

For instance, imagine we have a simple 1-D dataset where $y(t)=f(x(t)\;|\;\mu =0,\sigma ^{2}=0)$ for all t, where f is the Gaussian function. This could be perfectly modeled by a RBF with one center vector at 0 and appropriate β. However, if the x(t) are for some reason clustered far to the right, the center vector would be placed there, and the model would not work at all.

Or is the article talking about doing k-means clustering on the x and y dimensions together? — Preceding unsigned comment added by Allion (talk • contribs) 19:39, 18 November 2017 (UTC)