For Artificial Neural Network, I already tried many structures and the best result was used the number of hidden layer neurons same with the input layer neurons. Is there any special reason for that?
For the number of the hidden layer neurons, there is no general rule but empirical rules. The size of the hidden layer should be :
- equal to that of the input layer (Wierenga and Kluytmans, 1994)
- equal to 75% of it (Venugopal and Baets, 1994),
- equal to the square root of the product of the number of neurons in the input and output layer (Shepard, 1990).
Good information Nacera
https://www.researchgate.net/post/How_to_decide_the_number_of_hidden_layers_and_nodes_in_a_hidden_layer
Like Nacéra Benamrane already mentioned, there is no clear rule on this.
For example, a basic classification problem the number could be lower if an 'easy' seperation can be found during the training process. A higher number of hidden neurons could indicate that the problem is first needed to be transfered in a higher dimensional space where a solution is found.
All that is very dependent on your problem you're trying to solve and your data for training. Also neurol nets (especially more complex ones) are hard to analyse due to the amount of parameters and their black box nature.
- Badges
- Science topic