Is it rule for all segment or other kind of it?

{moderator edit - quiz question and answer removed}

What do you mean by “all segment or other kind of it”? The point is that there are different network architectures and there are other kinds of problems besides classifications problems. What activation function you use at the output layer will depend on whether it is a binary classifier (yes/no), a multiclass classifier (dog, cat, pigeon, aardvark, elephant, mongoose) or a regression problem (the price of a house or the temperature at 3pm tomorrow).

In neural network architectures, each “layer” typically looks like what happens in that picture: the first step is the “linear activation” which is a linear combination of the weights and inputs plus the bias. And then the second step is that you apply a non-linear “activation function” to the z value. In the particular case of a single layer perceptron, the \sigma function will be sigmoid, which is what is shown in the selected option. If you were doing a single layer regression, then it would be the squared error distance style loss function shown in options 1 and 3.

Please note that we aren’t supposed to publish quiz questions and answers or solutions to the coding assignments.