how many neurons should be in the hidden layer ?

Hi all How are you ? I have a question : I have to implement the Back Propagation algorithm that teaches the neuronal network to recognize hand written digits:{2,3,4,8} in the program should be only one hidden layer of neurons it's input 15*15 image of digit it's output is 4*1 vector of resemblance to each digit above BUT: How many neurons should be in the hidden layer ? I know that the amount of hidden layer neurons strongly affects the performance of the algorithm,so if anybody knows the answer please tell me !

Answer 1

[the cyclist]

I am not an expert in NN, but I know a little. One possibility, if you have enough data, would be to separate your data into a training set and a validation set. Then, train different networks with different numbers of neurons in the hidden layer (using the training set only), and see which one performs best on the validation set.

Answer 2

[Christos]

i agree with the cyclist. of course its not my field but in theory you have to test some networks and also try different activation functions in the hidden layer.