The ouput of Dense(1,activation='sigmoid'), is it the probability y=1 or y=0 for binary classification problem?

Question

I used the integer encoding for the classes 0 for first class and 1 for second. So just wanted to know the output corresponds to which probability first class or second class? Thanks a lot

Answer 1

There is no definition about the output. Anything is about how you encode your labels and the loss function.

A supervised NN tries to find weights for neurons to minimize the loss(difference) between the prediction(the output of the network) and the label(class you have). There are many different built-in loss functions in Keras or any other libraries/toolkits. For example, with the most common one MSE, the loss is like (output - class)^2. If the output of the network is 1, and the real label is 0, then the network tries to find parameters to minimize the difference 1 ((1-0)^2). From the example I gave, you can say the output 1 is for real label 1 and output 0 is for real lable 0. But there are other loss functions/or customized functions which are in opposite way. The key thing I want to say is the how you encode your labels and how you define/specify the loss function.

If you don't know the basis of NN, there are tons of good online resources.