I use Sigmoid activation function for neurons at output layer of my Multi-Layer Perceptron also, I use cross-entropy cost function. As I know when activation functions like Tanh is used in output layer it's necessary to divide outputs of output layer neurons by sum of them like what is done for softmax, is such thing necessary for sigmoid activation function? If it's necessary to normalize outputs of neurons, does it affect derivations?


Your Answer

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

Browse other questions tagged or ask your own question.