A classifier is said to perform online learning if it can detect novel class from the new input and register this class with classifier.

Since in most classifiers (particularly neural networks) we use softmax for the output function, how will the classifier detect a novel class? The outputs of the softmax sums to 1 so when a new novel item is fed to the classifier, what does it mean for the classifier to detect novelty?

It can't be that the softmax returns low probability for all the classes due to the sum to 1 constraint. Eventually, atleast 1 class will have the maximum class likelihood and assign this nov item to that class.



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.