The number of input neurons is equal to the output of the hidden layer 1. Then the number of outputs is equal to the number of predicted labels. We set a smaller value yet again, considering a very few inputs and features.
Here we used the Softmax activation function, which gives us a probability distribution over classes (the outputs sum to 1.0), and the losses function as cross-entropy for binary classification (XNET) since we want to convert the output (probability) to a discrete class, that is, zero or one:
OutputLayer output_layer = new OutputLayer.Builder(LossFunction.XENT) // XENT for Binary Classification .weightInit(WeightInit.XAVIER) .activation(Activation.SOFTMAX) .nIn(16).nOut(numOutputs) .build();