VQC question

We know that the VQC measures all qubits and for binary classification it then performs a parity functions to map the output probabilities to the class labels. But how does the VQC do this if there are more than two labels? The parity function only works for two classes and in the documentation i only found that the VQC uses some kind of Extension of the parity function. But thats it. We have no more information regarding this than „an extension of the parity function“. What is that extension and how does it perform the mapping between the probability vectors and the class labels?