I have a completely beginner question about the setup of artificial neural networks. Basically it boils down to:

  • How do I put in data and get results? For example, classification: let say I have 10 input features, all numerical, and 5 output classes.
  • How do I set up my input layer and my output layer? Would it be good to have 10 input nodes, one for each feature and feed each node a single numerical value during feed forward evaluation?
  • How about the output layer? Should I set up 1 node, 5 nodes?
  • What is the classification rule for neural networks?
  • $\begingroup$ I think you will have more luck with this question on the CrossValidated StackExchange, the statistics SE site. There are already quite a few NN questions, and a neural-net tag. $\endgroup$ – Michael Grant Apr 21 '13 at 1:32

From the top of my mind, or the rest of knowlegd about back-propagation neural networks, I would start with two layers with 10 perceptron in the first and 5 in the last using a the sigmoid logistic activation function.

$$ 1/(1+exp(-\beta x)) $$

Entering dimension 10, output dimension 5.

Better take a look at Haykin, S. (1999). Neural Networks: A comprehensive foundation (2nd ed). Prentice Hall International.

Chapter 4 Multlayer Perceptrons. If back-propagation NN most commom feed forward NN.

Also if you did not take a look, check this

If you give more details about your problem we might be able to give a better asnwer.

What you mean by classification rule?

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.