0

I would like to ask why features are weighted in machine learning before activation? Could you explain this on very simple and practical example?

Makaroniiii
  • 101
  • Do you mean in a neural network? – littleO May 29 '17 at 16:37
  • Yes littleO, I am complete newbie, I am looking for a kid friendly answers:) – Makaroniiii May 29 '17 at 17:01
  • This is not an answer to your question, but here are some tutorial videos I found recently that look like they might be good for beginners:https://www.youtube.com/channel/UCWN3xxRkmTPmbKwht9FuE5A/playlists – littleO May 29 '17 at 17:19
  • Thanks, I have seen it before, however I don't understand why data is being weighted still :) this is why i ask – Makaroniiii May 29 '17 at 17:21
  • Somebody else will have a better answer, but one possible answer is that neural networks were intended to loosely model what happens in the brain, and it turned out that neural networks work well so we have stuck with them. Neural networks are capable of approximating "any" function (in some sense) and they are simple enough that they can be trained efficiently (or at least, efficiently enough to be useful). I wonder myself if there are other families of parameterized functions that would be equally useful or more useful in machine learning. – littleO May 29 '17 at 17:31
  • What do you mean "weight the features"? Do you mean why is there a vector of weights associated to each neuron in a neural net? In general, features are not inherently weighted in machine learning; they are weighted by either a feature selection algorithm or by the learning algorithm itself (as part of the learning process, i.e. to learn the weights of the features is the purpose of the learning algorithm). One algorithm will assign different weights than others. – user3658307 Jun 05 '17 at 20:09

0 Answers0