I'm working on a neural networks project right now and for that I'm reading a bunch of scientific papers, in a few of those the terms additive hidden nodes and radial basis functions are thrown around, but I seem to have trouble to get a clear explanation of the terms there and anywhere else on the internet. I seem to have gathered that these are classifications for neuron types but I would love to get a more clear intuitive explanation of the terms. Preferably one that doesn't require me to be very mathy to understand. I'm fairly new to neural networks so beyond sigmoid neurons and backpropagation algorithms I'm still a bit lost when it comes to the common termonology.
$\begingroup$
$\endgroup$
2
-
$\begingroup$ Are you reading Huang's ELM papers? $\endgroup$– rcpintoJun 8, 2016 at 22:28
-
$\begingroup$ No, but work based on it (though I should really read that one as well) $\endgroup$– Hasse IonaJun 9, 2016 at 12:16
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
0
Albeit not wrong, Huang seems to be the only person in the world to use the term "additive hidden nodes". By this, he means that the neuron computes the sum of weighted inputs. In other words, the kind of neural network you're already used to.
An RBF neuron, on the other hand, computes a distance (usually the Euclidean distance) from input to some center (which can be thought as the weights if you see them as a vector) and applies a exp(-dist²) function in order to obtain a Gaussian activation. Thus, RBF neurons have maximum activation when the center/weights are equal to the inputs.
